Thread by @DrDamluji

Prognostic Value of Machine-Learning- Based PRAISE Score for Ischemic and Bleeding Events in Patients With Acute Coronary Syndrome Undergoing PCI: @JAHA_AHA

Editorializing: 3 major applications of AI/ML in medicine: automation, prediction, cognitive reasoning!

Summary 👇👇

1/📊This study investigates PRAISE score for predicting 1-year all-cause mortality, recurrent AMI, and major bleeding events in post-PCI ACS patients treated with DAPT.

Application for #prediction in medicine using 6433 ACS pts

2/🔍Study finds PRAISE score has moderate discrimination for all-cause mortality but poor discrimination for recurrent AMI and major bleeding in an Asian population.

(Chinese cohort)

3/📈PRAISE score has similar discriminative capacity to GRACE 2.0 for all-cause mortality, but lower specificity.

4/🤝PRAISE score was similar to other scores for predicting recurrent AMI and major bleeding events.

PARIS, Grace, and Precise DAPT

Good start - this is the beginning of improved prediction using unexplainable deep learning models with large # of features

5/❌PRAISE score overestimates risk for all 3 endpoints and has relatively poor calibration in Asian patients with ACS.

6/🩺Decision curve analysis (DCA) shows PRAISE score is useful comparable to GRACE 2.0 for all-cause mortality if the 1-year risk of 5-10% is considered as a relevant threshold.

7/💻Machine learning approaches enable improved prediction using large data, but currently may be insufficient for meaningful clinical changes.

This will change: Using TIMI and Grace in my Cath reports will be outdated in few years.

8/📚Literature on using classification algorithms for various complexity-enhanced predictions in medical disciplines will grow.

9/👥Limitations include differences in characteristics of validation cohort and unmeasured baseline characteristics.

single center

external validity and others

but great attempt

10/🧐In Conclusion: PRAISE was able to predict 1 year mortality, but it over predicted risk in high risk patients, and for bleeding, and recurrent AMI.

AL/ML will have a role in medicine and will change how we use data, particularly in light of wide adoption of EMR in the US.

Bravo @JAHA_AHA