Background And Aims

Personal prediction of stroke recurrence based on modifiable factors may enhance patient compliance to treatments and healthy lifestyles while optimizing resources of health centers. Vascular risk factors, socio-economic indicators and habits correlate with stroke morbidity and further recurrence; however, individual risk prediction is not accurate. We aimed to estimate stroke recurrence probability as a function of time (3months, <1 year and >1year), both at individual level and with larger classes of individuals.

Methods

Clinical and socioeconomic public healthcare-based dataset of 41325 patients admitted with stroke diagnosis in 88 public hospitals over 6 years were analyzed.

Results

Overall, 8509 patients presented a stroke (one or more) recurrence (20.6%), with the following temporal distribution: 3 months, 1year and >1year, 57%, 17% and 26% respectively.

We developed a supervised-machine learning based study and identified modifiable and non-modifiable risk factors with stronger impact on risk of stroke recurrence. An algorithm able to provide individualized risk of stroke recurrence at 3 and 12 months was developed (AUROC = 0.71). The risk can be continuously updated according to the status of modifiable risk factors.

We also calculated the survival curve for each patient, to detect different risk recurrence periods along time

Conclusions

Machine learning analyses can improve risk prediction and offer individualized information to patients that can be used as feedback for secondary prevention strategies. Our results pave the way to scalable Artificial Intelligence assisted care.