Speakers

Isabella Friis Jørgensen

PRESENTATION

Assistant Professor

University of Copenhagen

Novo Nordisk Foundation

Center for Protein Research

Comorbidity networks and disease trajectories from population-wide registry data

Abstract

Advances in medical science and modern medicine have increased life expectancy and thus, multimorbidity, when one patient is diagnosed with more than one chronic disease, is an increasing challenge. The temporal patterns of diseases over the life course reflect underlying biological mechanisms, shared causal factors, or appear because of other conditions or their treatment. We investigate disease comorbidities using a previously published method to identify frequent disease trajectories comprising time-ordered comorbidities of which many present in a temporal order because of underlying genetic and environmental factors. We have used the Danish National Patient Registry (NPR), which contains hospital diagnoses for more than 7 million patients for the entire Danish population. During the talk I’ll give examples of different types of visualizations of disease trajectories using networks approaches including the Danish Disease Trajectory Browser (dtb.cpr.ku.dk).

I will also give a more recent example of how disease trajectories can be used to discover patients at risk of mis- and over-diagnosis. The case is exemplified using Chronic Obstructive Pulmonary Disease (COPD) for which literature suggests that 5-60% of cases are mis- or over-diagnosed. We created temporal disease trajectories of all COPD patients which illustrate common disease paths before and after the COPD diagnose. Subsequently, we identified COPD patients whose individual disease history show a low similarity to the typical trajectories and thereby the COPD diagnosis appears in an unusual context. One subgroup comprising 2,185 patients at risk of being misdiagnosed with COPD as it seems likely that they are underdiagnosed with lung cancer as their laboratory test values and survival pattern are more like lung cancer patients than other COPD patients. Furthermore, only 4% had a lung function test registered in NPR to confirm the COPD diagnosis. Another subgroup with 2,368 patients was found to be at risk of “classically” over-diagnosed COPD as they survive more than 5.5 years after the COPD diagnosis, but without the typical complications of COPD. To the best of our knowledge, this is the first method to systematically identify patients at risk of mis- and over-diagnosis and it could be used in a more real-time clinical setting by discovering patients with unusual disease patterns. Subsequently clinicians could verify their COPD diagnosis with spirometry and consider other differential diagnoses more thoroughly.

Alexia Giannoula

Senior Researcher

Hospital del Mar Medical Research Institute (IMIM)/Pompeu Fabra University (UPF), Barcelona

Identifying temporal patterns in healthcare service-use trajectories of long-term breast cancer survivors

Long-term breast-cancer survivors (BCS) constitute a complex group of patients due to their increased risk of suffering long-term treatment side effects, age-related comorbid diseases and other negative outcomes. Given the estimation for their number to continue rising, the medical society has manifested the necessity for a dedicated long-term clinical follow-up. Using a large retrospective longitudinal cohort of Spain, a data-mining methodology based on unsupervised clustering and Dynamic Time Warping, is presented for the comprehensive exploration of BCS’ trajectories of their use of healthcare services over a 5-year follow-up. Applying data mining, for the first time, on this group of patients and incorporating, in addition, the temporal dimension permits the identification of complex temporal service-use patterns, that otherwise would remain undiscovered. These are next represented by condensed descriptive network diagrams and aggregated time/patient information to facilitate comparison between cases and controls. The patterns reflecting the use of radiology and hospital admission are particularly examined, in which a more intense and complex use made by the BCS, as opposed to controls, is revealed and analyzed in depth. The presented methodology could serve as the basis for better understanding the BCS’ needs, in order to more efficiently predict their forthcoming challenges and ultimately, improve their healthcare management.

Jana Lasser

PRESENTATION

PostDoc

Computational Social Science lab,

Graz University of Technology and

Complexity Science Hub Vienna

Stress-testing the Resilience of the Austrian Healthcare System Using Agent-Based Simulation

Patients do not access physicians at random but rather via naturally emerging networks of patient flows between them. As mass quarantines, absences due to sickness, or other shocks thin out these networks, the system might be pushed to a tipping point where it loses its ability to deliver care. Here, we propose a data-driven framework to quantify regional resilience to such shocks via an agent-based model. For each region and medical specialty we construct patient-sharing networks and stress-test these by removing physicians. This allows us to measure regional resilience indicators describing how many physicians can be removed before patients will not be treated anymore. Our model could therefore enable health authorities to rapidly identify bottlenecks in access to care. Here, we show that regions and medical specialties differ substantially in their resilience and that these systemic differences can be related to indicators for individual physicians by quantifying their risk and benefit to the system.

Peter Klimek

Associate Professor

Medical University of Vienna and

Complexity Science Hub Vienna

Generalized disease network models for the prediction of multimorbidity

Many countries face increasing life expectancy in their population. Health states of elderly patients are typically characterized by the cooccurrence of multiple chronic diseases, so-called multimorbidity. Yet, the trajectories along which patients accumulate disorders as they age are not yet fully understood. In this talk I discuss our recent efforts to understand the development of multimorbidity in the Austrian population by means of generalized disease networks. We base our analyses on a comprehensive dataset of all hospital stays in Austria between 1997 and 2014. Generalized disease networks are constructed from this data by identifying clusters of diagnoses that frequently cooccur in patients. As individuals age and acquire new diagnoses in hospitals, their cluster membership might change, allowing the empirical estimation of transition probabilities between different diagnosis clusters. We discuss the structure of this generalized disease network and unravel network regions that can be related to trajectories of healthy or unhealthy aging, i.e., regions of higher or lower mortality due to cardiovascular diseases. We further develop a compartmental model to predict the future development of multimorbidity as well as how certain shocks might impact the distribution of diseases in the population. We use this model to study expected impacts of post-acute cardiovascular and neurological sequelae of SARS-CoV-2 infections in Austria. Our work shows how population-scale models informed by administrative claims data can be used to more precisely identify vulnerable groups in the population and to proactively anticipate long-term impacts on public health due to external shocks like the recent COVID pandemic.