Local Early Warning System for Control of Infection Outbreaks
Project content and objectives
With the emergence of infectious pathogens with pandemic potential, such as the coronavirus pandemic caused by SARS-CoV-1 in 2002/2003 or avian influenza A/H5N1 in 2005, national pandemic plans for the health care system have been developed in Germany. However, a control system for infectious disease outbreaks that helps health departments develop individual goals and effective measures to manage a pandemic under local conditions has not yet been realized. The experience of the SARS-CoV-2 pandemic demonstrates the need for a practical and user-friendly platform for monitoring local outbreak events, taking into account specific regional structures.
Therefore, the aim of the LOKI-pandemics project is to develop a platform that allows public health authorities to control local infection events by introducing customized non-pharmaceutical measures.
In order to be able to detect and evaluate outbreaks at an early stage on a local level, the project analyzes infection-relevant data using various methods ranging from artificial intelligence to epidemiological modeling. The platform offers the possibility to interactively simulate different scenarios. In doing so, employees of public health departments can have different scenarios with and without measures displayed in order to map the influence of measures on the incidence of infection, to view the further development of outbreaks and to develop possible containment strategies.
LOKI-pandemics consists of four subprojects.
Subproject 1 is concerned with the collection and processing of data from various sources. These are relevant for use in the subsequent subprojects and form the basis for the development of LOKI and the underlying models. The goal is to integrate multiple data sources into one data pool for models and forecasts in a structured and easily usable way. This includes data collection, structuring, and primary analysis. In addition, the potential for integrating new data sources is analyzed. These data are either publicly available or are generated through collaborations with the pilot health departments. Furthermore, Subproject 1 addresses privacy issues. A major limitation in assessing and predicting epidemic dynamics is the availability, quality, and origin of data. Integrating a variety of independent data sources, standardizing them, and automating their processing are essential for downstream statistical analysis and accurate predictions. Quality control mechanisms are established to ensure high accuracy of data and metadata with minimal manual maintenance. At the same time, data is aggregated, combined, and converted into standardized formats to ensure a reliable and predictable data source for automated workflows.
Epidemiological models must be calibrated for predictions. However, reliable estimation of model parameters from insufficient data is often impossible. This problem is exacerbated as the level of model detail increases. Directly measurable model parameters derived from data provide valuable information, and their use improves the reliability of predictions.
Primary data sources include aggregated public surveillance data, case-based data from local health authorities, data on clinical and hospital parameters (in collaboration with the Lean European Open Survey on SARS-CoV-2 [LEOSS] project, a prospective European multicenter cohort study), population-based studies, and seroprevalence surveys (https://serohub.net/en/, https://hzi-c19-antikoerperstudie.de/).
In addition, Subproject 1 integrates existing data on vaccination, wastewater sampling, contact networks, mobility, and aerosol dynamics.
In addition, an evidence synthesis of existing literature will be conducted to integrate previous estimates of parameters. The evidence synthesis will focus on SARS-CoV-2 in the first phase and later integrate ongoing work (https://respinow.de/) on other respiratory pathogens.
These data and the synthesized evidence on model parameters will be made available in a structured form for downstream subprojects to use directly in pattern recognition approaches, mechanical models, and in model calibration and uncertainty analysis. To promote synergies with existing international projects, Subproject 1 will seek to incorporate and compare the current platform in an overarching review of similar global projects to build modeling and prediction platforms for use by public health agencies.
Several areas of the subproject deal with sensitive data. The data collected will be analyzed for its potential to violate privacy and the risk from inferences will be examined. Similarly, it will be ensured that the products of data analysis or algorithms do not expose private data and are not vulnerable to inference attacks. Where necessary, local mechanisms will be deployed and developed. Data will be aggregated and analysis coarsened to ensure individual privacy is protected.
Aggregation of a variety of data sources will provide a detailed picture of the pandemic and its parameters, allowing unprecedented precision of prediction in a thoughtful, semi-automated manner accessible to public health workers and other relevant professionals.
The goal is to integrate multiple data sources into one data pool for models and forecasts in a structured and easily usable form (for use in the following subprojects).
Subproject 2 “Analysis” deals with the development of models that are intended to represent the real spread of infectious diseases and provides methods for parameter estimation, uncertainty quantification and optimization. Parameter estimation attempts to draw conclusions about the real values of various parameters (e.g., the number of daily contacts). Uncertainty quantification is used to calculate how likely certain results are if certain aspects of the model are not fully known or are fuzzy or subject to variation. Optimization is used to achieve the best possible results according to a predefined goal.
Overall, different types of models are used. The first class of models ranges from classical SIR-type models based on ordinary differential equations to models involving integro-differential equations and coupled regional or metapopulation models. In addition, LOKI-pandemics will provide agent-based models as well as models based on classical statistics and machine learning. To explore infection dynamics, these models will be used either independently or with methods for parameter estimation and optimization. To ensure that the models can be used for future epidemiologically relevant outbreaks, the scientists in subproject 2 will also establish a workflow for calibrating and integrating real-time data into the models.
Subproject 2 will provide retrospective analyses of the Sars-CoV-2 pandemic to draw conclusions from past outbreak developments. These inferences will help increase the depth of infectious disease spread models or, if needed, develop new models for potential future respiratory viral disease outbreaks. In addition, retrospective analyses will help to evaluate the impact of regionally adopted non-pharmaceutical measures, such as those implemented in Germany, in order to use these findings for future outbreak containment.
All models are implemented efficiently and in a modular fashion so that they can provide real-time results in the form of visualizations of the further development of an outbreak event and are easily interchangeable in the near future as newer models are developed. Regionally resolved models can also be used to incorporate regional aspects of disease spread or control. Subproject 2 will also allow simulation of scenarios with and without non-pharmaceutical interventions. This will allow the impact of these interventions on the incidence of infections or hospitalizations to be illustrated. Taking into account uncertainty quantification, the results will help to evaluate outbreak events and support the decision to introduce non-pharmaceutical measures and their optimization. Numerical optimization methods, among others, are used to recommend adapted measures.
The workflow of subproject 2 will be designed to be as general as possible, so that transfer to new pathogens will help to reduce the costs of preparing for new pandemics. The lessons learned from Sars-CoV-2 in the subproject will thus enable decision-makers and the German population to be better prepared for potentially upcoming pandemics.
The goal of subproject 2 is to provide modular and efficiently implemented models for the spread of infectious diseases as well as methods for parameter estimation, uncertainty quantification and optimization. In addition, a workflow that is as automatic and general as possible will be created to enable the computations of different scenarios of infectious disease spread.
Subproject 3 comprises the technical implementation and realization of the platform. The modular components developed in subproject 2 are combined here to form a common platform. This is based on a cloud-based infrastructure with direct connection to a supercomputer for computationally intensive operations and access to large storage capacity. In addition to the computation of user-driven simulation scenarios, visualization and visual analytics methods are an essential aspect of the LOKI platform.
Therefore, at the heart of the platform is the web application with interactive components that visualize and provide easy access to the model results. The collected data and the results of data analyses, simulations and forecasts are communicated to the target audience in a concise, pragmatic and understandable way.
By combining visualization with interaction and automated algorithms, the web application supports decision makers in evaluating possible policy measures. The algorithms simplify the classification of complex phenomena and the detection of new patterns in the data. Diagrams and graphs can be displayed in detail and modified interactively.
Seamless integration with existing systems is critical for a realistic, cost- and time-efficient implementation of the platform and adaptation to targeted user groups. A dedicated API will allow other systems to gain access to the platform’s key results and aggregated and simulated data to facilitate the reuse of our algorithms and data in other systems. There will also be an active effort in the opposite direction to leverage and technically integrate additional data sources.
Information security aspects are an integral part of the overall platform to ensure that sensitive data is analyzed while maintaining privacy. Solutions will be defined primarily by technical and legal considerations, which will require collaboration with legal experts and will affect the design of the platform’s security architecture. Because the platform will provide various forms of access to potentially sensitive data, it is critical to develop novel security architectures that can balance data use with privacy and security considerations.
The ambitious goal of an integrated platform for different methods of data analysis, data storage and user interaction requires modularity and flexibility in the software architecture. Only then will the generated platform serve pandemic control beyond the project lifetime and be generalizable to different respiratory infections.
The goal of this subproject is the technical implementation and realization of the platform with the development of the web application for the users as the central component. From here, user-controlled simulation scenarios can be initiated and results visualized and analyzed. Interfaces for the integration of the platform into existing systems, the coupling to external data sources and a high demand on information security make the platform open and flexible to use.
Transfer into practice
Subproject 4 is responsible for the transfer of the platform into practice. This means that this subproject and the responsible employees form the human interface between the developers of the platform and the cooperating pilot health offices.
To fulfill this function, this subproject organizes and structures the exchange and communication between the development and the application of LOKI pandemics.
By involving users in the development process of the platform, valuable experiences and suggestions from the field can be directly incorporated into the further development of the application. This makes it possible to adapt the platform to the individual needs of a health department. Various formats are used to establish and coordinate communication between development and application. Regular web conferences, for example, ensure direct contact with the developers and enable the exchange of experiences between participating pilot health departments.
Following the launch of the platform, subproject 4 staff will train and advise pilot health departments. The trainings will be offered in different formats. Furthermore, training materials will be developed and made available to the pilot health offices.
Another task of subproject 4 is to evaluate the platform. The aim is to demonstrate the benefits and quality of the application for daily practice. The staff can use the feedback from the users during the development process to evaluate and optimize the platform.
The aforementioned tasks of the subproject are supplemented by the provision of user support to enable personal exchange with the pilot offices and rapid clarification of open questions. By means of a telephone hotline, the employees of subproject 4 are regularly available for questions, ideas and suggestions.
The aim of subproject 4 is to design the platform in a user-friendly and practice-oriented manner in cooperation with the health authorities and the developers.
The focus is on ensuring the transfer of the platform into the practice of the public health departments. A successful transfer is to be achieved through training and support possibilities, but also through the coordination of exchange forums.
Project responsibles and partners
Prof. Dr. Michael Meyer-Hermann
Helmholtz Centre for Infection Research
Dr. Martin Kühn
(deputy project leader, coordinator SP2)
German Aerospace Center
Dr. Berit Lange
Helmholtz Center for Infection Research
Jens Henrik Göbbert
Akademie für Öffentliches Gesundheitswesen