138 Investigating overdiagnosis using temporal comorbidities patterns. (20th August 2018)

• Record Type: Journal Article
• Authors: Jørgensen, Isabella Friis; Brunak, Søren
• Is Part Of: BMJ evidence-based medicine. Volume 23:Supplement 2(2018)
• Page Start: A65
• Page End: A65
• Publication Date: 2018-08-20
• Subjects: Evidence-based medicine -- Periodicals; 616.005
• DOI: 10.1136/bmjebm-2018-111070.138
• Languages: English
• ISSNs: 2515-446X

Abstract

Abstract : Objectives: During the last decades the prevalence of medical conditions has increased, thus more and more patients are living with numerous diagnoses of which many represent comorbidities that appear as a consequence of the primary condition. Overlapping symptoms in these comorbid conditions complicate the process of making the right diagnosis. Thus, missed diagnoses, misinterpretation of symptoms and overdiagnosis are common problems when diagnosing patients with multiple morbidities. Here, our objective was to identify flawed diagnoses by exploring the co-occurrence patterns among comorbidities and investigating significant time-dependent disease correlations using population-wide data from Denmark. Methods: This study takes advantage of a population-wide disease registry, The Danish National Patient Registry, which covers all hospital encounters in Denmark from January 1994 to December 2015 and includes 6.9 million individuals. Many patients share many chronic diseases and their associated comorbidity patterns and the concept of temporal trajectories of diagnoses can be used to pinpoint diagnoses, which appear unusually in the context of other diagnoses. Statistical significant trajectories were created for the entire population to identify common disease paths and diseases, which might be overdiagnosed. Results: Investigating temporal diagnose correlations of the 6.9 million patients results in 13 436 and 684 significant trajectories consisting of four or five consecutive diseases, respectively. We devised an 'inverted imputation' scheme that identifies diagnoses which appear in unusual contexts, and observed that the diagnoses of chronic obstructive pulmonary disease, asthma and several cancers stood out as having frequently observed contextual diagnoses, which at the same time, relative to the overall frequency, appear in other contexts. The scheme allowed us to pinpoint individuals with these types of diagnoses that may reflect overdiagnoses that appear without the most common comorbidity context. Conclusion: We show that population-wide data covering long time periods can be used to support analyses of temporal disease trajectories that point at specific individuals that may be overdiagnosed. In Denmark other types of data, including biochemical test values, medication patterns and socio-economic data can be linked to these individuals for further substantiation of the characteristics of different overdiagnosis aspects.