Data clustering shows comorbidity networks
Diseases often appear in clusters with other diagnoses. Clinicians call that “comorbidities.” One could also speak of comorbidity networks. Take obesity as an example. Severely overweight people (diagnosed with adipositas) often suffer from a whole set of other conditions too, including metabolic problems (pre-diabetic or diabetic), cardiovascular diseases, muscosceletal problems, or depressions.
However, one could think that people could be obese and still perfectly healthy. And if that was the case, under what circumstances are the extra pounds then rightly called pathologic? Is there maybe more than one “obesity”? Are there maybe more types of diabetes than we know of? Or could a totally different description of a disease be more appropriate than the definition that is currently used?
A team of CSH researchers investigated if big medical data could produce new, maybe even surprising insights into comorbidities.
During his time at the Hub, our former researcher Markus Strauss in collaboration with Stefan Thurner and Peter Klimek worked on a method that automatically identifies disease phenotypes in a large medical dataset. Disease phenotypes are based on different comorbidity contexts of diagnoses rather than on single medical definitions.
The new method, published today in the Journal of the Royal Society Interface, allows to identify distinct disease phenotypes made up of clusters of diagnosis features. The scientists think these phenotypes might correspond to different disease aetiologies that cannot be captured by single diagnoses and their interactions.
Complex diseases, they write, have in common that they (a) cannot be captured by one single specific diagnosis (for instance as “metabolic syndrome”) and (b) that a specific diagnosis can present itself as part of multiple disease phenotypes (e.g. hypertension).
The new approach allows for the first time to investigate multiple roles of specific diagnoses. In particular, “it enables us to discover and analyze novel disease phenotypes based on different comorbidity contexts of diagnoses,” says Markus. This differentiation has strong implications for a better understanding not only of disease aetiology but also for the meaning of a specific given diagnosis in terms of treatment and prognosis.
To validate their method, the scientists used an anonymized dataset, consisting of 45 million recorded hospital stays of 9 million patients in Austria over a time period of 18 years (1 January 1997 to 31 December 2014).
Healthy and unhealthy obesity
The paper shows for the first time that specific features of obesity are part of different clusters on the comorbidity network. In the scientific literature, the concept as well as the usefulness of distinguishing people with “metabolically healthy obesity” (MHO) from people with “metabolically unhealthy obesity” (MUHO) is still under debate. According to the current study, this distinction seems to make sense.
The researchers found an obesity phenotype with low prevalence of diabetes and other metabolic disorders (corresponding to MHO), and a phenotype where hypertension, hyperlipidaemia, and diabetes often co-occur and where the risk of developing new metabolic and cardiovascular risk factors is considerably higher (MUHO). Although people with MUHO clearly outnumber the MHO’s, people with overweight but without other health conditions should maybe be treated differently in the future.
However, adds Peter, “over longer time spans metabolically healthy obese people also show a slow progression to a more unhealthy type of obesity, suggesting that the MHO type cannot be considered completely benign in terms of prognoses, but lies somewhere between MUHO and non-obese persons.”
Another finding was that not only the risk of having a metabolic disorder was significantly higher in people with MUHO, compared to the healthier obese; but also the prevalence of mental disorders. “It can be speculated that shared psychosocial factors are underlying pathophysiological mechanisms starting a vicious circle between unhealthy lifestyle, metabolic and psychological disturbances, which again are linked and mediated by obesity,” says the paper, which was co-authored by the endocrinologist and gender medicine expert Alexandra Kautzky-Willer and Thomas Niederkrotenthaler, both Medical University of Vienna (Thomas is also member of the CSH Associate Faculty) .
Altogether, a data-driven approach to open questions in medicine proved to be helpful once again.
“This elaborated example on obesity demonstrates how to use the developed method to obtain meaningful novel information about complex diseases. Different from clinical studies in the topic area, which typically follow obese and non-obese patients over time, we show for the first time, that the disease phenotypes of obesity identified in this purely data-driven approach are somewhat consistent with the discussed clinical phenotypes of obesity,” the authors conclude.
Markus Strauss, Thomas Niederkrotenthaler, Stefan Thurner, Alexandra Kautzky-Willer, Peter Klimek, Data-driven identification of complex disease phenotypes, Journal of the Royal Society Interface 18 (180) (2021) 20201040
Una red global de científicos medirá qué éxito tuvieron las medidas sanitarias para frenar la pandemia [Spanish]
Infobae, Sep 16, 2021
V. Machàcek, M. Srholec, M. R. Ferreira, et al.
Researchers’ institutional mobility: bibliometric evidence on academic inbreeding and internationalization
Science and Public Policy (2021) scab064
M. Hâncean, J. Lerner, M. Perc, M. Ghita, D. Bunaciu, A. Stoica, B. Mihaila
The role of age in the spreading of COVID-19 across a social network in Bucharest
Journal of Complex Networks 9 (4) (2021) cnab026
J. Einsiedler, et al.
Transferable Models to Understand the Impact of Lockdown Measures on Local Air Quality
in: Proceedings of the 3rd International Workshop on Urban Computing (UrbCom) (2021)
M. Hakkar,et al.
Enhancing rare earth element transfer from phosphate rock to phosphoric acid using an inexpensive fly ash additive
Minerals Engineering 172 (2021) 107166