Background and Aims: Cardiometabolic comorbidities frequently develop in individuals with hypertension, type 2 diabetes, or hypercholesterolemia. When present in combination, the risks of vascular disease are additively increased.

Methods: We created an artificial intelligence neural network to forecast the development of additional comorbidities when a single cardiometabolic disorder was present. The neural network was first taught diagnostic criteria to reproduce current clinical guidelines with 95% accuracy and, in a second step, trained on the outcome of anonymized electronic medical records of 48225 adults from South London primary care clinics, where 6127 had hypertension, 3001 had type 2 diabetes, and 5630 were treated for hypercholesterolemia. Among these cohorts, 2142, 1482 and 724 developed a second morbidity within a 3 year follow up period.

Results: Prediction of an initial comorbidity had a sensitivity of 0.94±0.14, 0.85±0.32, 0.83±0.36 and specificity of 0.87±0.24, 0.90±0.19, 0.93±0.14 up to 1, 2 and 3 years prior to the diagnosis respectively. A second comorbidity was then predicted with a sensitivity of 0.63±0.1, 0.63±0.1, 0.52±0.21 for patients with hypertension who developed type 2 diabetes and 0.76±0.09, 0.70±0.08, 0.85±0.13 for patients with type 2 diabetes who developed hypertension.

Interpretation: This proof-of-concept analysis demonstrates the utility of a neural network to predict the development of one or more comorbidities based on data routinely collected in primary care. At the level of individual patients knowledge of the probability of acquiring additional comorbidities could provide opportunities for prevention. Non-pharmacological and specific pharmacological interventions could be directed at avoiding or postponing additional cardiometabolic risk factors.

Disclosure

A.Krentz: None. L.S.Brunschwig: Employee; Metadvice SA. Y.Dibner: Employee; Metadvice. H.Michel: None. A.Jaun: Board Member; Metadvice.

Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.