Skip to main content

Sparse Composite Likelihood Inference in Count Time Series

Periodic Reporting for period 1 - SCouT (Sparse Composite Likelihood Inference in Count Time Series)

Reporting period: 2016-06-01 to 2018-05-31

The analysis of infectious disease data is essential for strengthening surveillance systems and supporting the response to public health threats. Statistical modelling of infectious disease data needs to account for a variety of aspects including seasonality, trends, the effect of concomitant infectious diseases, human activities that may favor the disease diffusion and external factors such as weather conditions. Infectious disease data typically consist of numbers of cases that are recorded sequentially in time, usually on a daily or monthly basis, and present some temporal dependence, e.g. the evolution of a disease today depends on its evolution the day before. On the other hand, monitoring of a single disease might be of limited interest for outbreak detection purposes, especially in situations where the same underlying process, such as a common pathogen, has generated multiple disease series introducing dependencies between them. In such cases, there is an emerging need for multivariate surveillance techniques that can account for such dynamics and thus improve the effectiveness and timeliness in outbreak detection. Failure to account properly for these data features can result in incorrect statistical results.
Nowadays, a variety of sophisticated statistical methods able to account for the characteristics of infectious disease data is available in the literature. However, the computational difficulties that often come along with the implementation of these methods, limit their practical usefulness. The SCouT project copes with methodological and implementation aspects of advanced statistical techniques aiming to enhance their flexibility and applicability for health surveillance purposes including identification of disease outbreaks. An example of the application of the developed methodology to the analysis of infectious disease data is displayed in Figure 1. The figure shows the monthly number of invasive meningococcal disease cases in Greece for the years 1999-2016 as obtained by the European Center of Disease Control (ECDC) Surveillance Atlas. Appropriate statistical modeling of such data allows for an effective prediction of the number of future disease cases and thus identification of disease outbreaks. For example, modeling of the invasive meningococcal disease data available until the year 2015 allows us to predict the 2016 cases and identify disease outbreaks, i.e. points outside the shaded area in Figure 1.
The statistical methods developed in line with the research objectives of the SCouT project enrich the available toolbox for public health surveillance and will hopefully contribute in the effective and timely detection of disease outbreaks.
To successfully meet the objectives of the SCouT project a number of activities have been carried out during the project lifespan. Most of these activities were foreseen by the five work packages of the SCouT work plan, namely WP1-Management, WP2-Training, WP3-Methodological Research Work, WP4-Applied Research Work and WP5-Communication and Dissemination. Management activities mainly consisted of administrative and financial management of the project and were carried out with the support of the host institution. Training activities included creation of a career development plan, hands-on training and directed reading of the literature, participation in the master-level courses “Numerical algorithms” and “Network Science” at the host institution, attendance of the European R Users Meeting 2018 and a specialist course at the University of Padova for developing skills in efficient scientific programming and high-performance computing. Additionally, the fellow held a seminar on time series models at the University of Padova and she participated in training sessions on European project design organized by the host institution and in European IPR training activities dedicated to IPR management with focus on the Marie Curie Actions. She also attended Italian language courses at the Ca’ Foscari School for International Education. The methodological research work was implemented in collaboration with the supervisor, Dr. Cristiano Varin, and a broad research network in order to meet the technical aspects of the project that is development of the statistical methodology and related algorithms for its efficient implementation. The established research network of the SCouT project includes international collaborations with the Technical University of Dortmund (partner organization), Athens University of Economics and Business and University of East Anglia. The applied research work consisted of the application of the developed methodology on infectious disease data, partly sourced by the European Center of Disease Control (ECDC) Surveillance Atlas. Training and research activities were also implemented during three secondment visits of the fellow to the Technical University of Dortmund and a short visit of the secondment supervisor, Prof. Fried, to the University of Venice. Communication and dissemination activities included development of the project’s public website, participation of the fellow as speaker in the MSC Info Day 2017 organized by the Ca’ Foscari Research Office and APRE, oral presentations in three renowned international conferences, two published conference papers, two scientific papers already submitted for publication in academic journals of high reputation, several other papers being in progress and/ or close to submission and development of a dedicated open source software (R package). Other activities not foreseen by the SCouT work plan regard the establishment of multidisciplinary collaborations with other scientists from University of Venice in the framework of the ERC funded project “Early Human Impact” ( Although these activities are not intimately linked to the SCouT project, they have broadened the fellow’s knowledge of time series methods and their applicability, contributed in her integration with the scientific community of the University of Venice and resulted in two published papers and another paper close to submission.
The methodology developed in the framework of the SCouT project has multiple contributions in both the count time series literature and in public health surveillance problems. Firstly, it addresses a number of methodological and technical issues related to the estimation and fitting of time series models for counts. Secondly, it highlights the usefulness of count time series models in the statistical analysis of infectious disease data and suggests additional modelling options in this framework. Finally, it provides an open source software that can facilitate the application of the suggested methodology to a wide variety of problems. With regard to public health surveillance that motivates the research work of the SCouT project, the developed methodology is expected to contribute in the early warning and detection of emerging threats and the development of disease monitoring across Europe.
Figure 1: Application of the suggested methodology on public health surveillance data