Autore: Linh Hoang Khanh DANG, University of Bologna
linhhk.dang@unibo.it
Nicola Caranci, Rosella Rettaroli, Giulia Roli, Daniela Fortuna, Elena Berti, Rossella Miglio
Categoria Primaria: Malattie cronico-degenerative e tumori
Categoria Secondaria: Metodi epidemiologici e biostatistici
Introduction. One key factor to construct appropriate and sustainable healthcare plans for aging populations is the possibility of identifying precise multimorbidity patterns and seizing their progression in time. Traditional approaches like factorial analysis often results in diseases being included in several groups at the same time, making interpretation of resulted patterns rather complicated. Meanwhile, even if cluster analysis allows data partitioning into separate groups, it cannot integrate the association strength between diseases within a cluster. In this context, the combination of probabilistic approach from graphical model and intuitive visualization from network analysis is emerging quickly as powerful tool in recent years to not only efficiently explore the richness of administrative health data, but also to provide an analysis framework whom predictability can be assessed.
Object. We identify the multimorbidity patterns using graphical models combined with network analysis in a population aged 50 and older in Emilia-Romagna region in 2011 and followed up to 2019, before studying the profiles of patients within identified patterns.
Method. Emilia-Romagna Longitudinal Study was integrated with micro-data on more than 30 chronic diseases prevalences resulted from algorithms applied on regional health databases (2011-2019). We selected a closed cohort of residents at 2011 census stratified by sex and age (50-64 and 65+). The chronic diseases network was estimated using mixed graphical model, and the diseases’ prominence within the network was assessed using various centrality measures (e.g., the “gatekeeper” diseases in pattern formulation based on betweenness centrality). We applied the most suitable community detection method for our data based on modularity score and identified the clusters within the established networks. Descriptive analyses were performed to characterize the individuals in each pattern.
Results. Among the 1,010,571 individuals (557,562 females, 453,009 males) with at least two chronic diseases, the multimorbidity patterns and their diseases composition vary across population strata. Even if heart failure and dementia are not the most prevalent for younger females (2.5% and 1.3% respectively), they still play important role as gatekeepers in morbidity accumulation. Similar result is also found for older females. For younger and older males, cirrhosis holds ineligible place despite a prevalence of 3%. And regardless of ages group and gender, patients with multimorbidity aged 50 and above primarily belong to the pattern driven by cardiovascular diseases.
Conclusion. Graphical model combined with network analysis offer a global view of the interconnected chronic diseases in a population, both in cross-sectional and longitudinal data framework. Insights from this research line could help develop appropriate diseases prevention and healthcare resources allocation tailored to the target population.