759-3 Prediction Models for Outcome of Coronary Surgery and Coronary Angioplasty on a Desktop

Abstract

There is a need in clinical practice for rapid accessibility of outcome data in rapid fashion for clinical decision making. Data for this purpose come largely from clinical trials and databases. Emory University has a cardiovascular database with 80,000 patients which includes all cardiovascular surgical procedures and interventional procedures in the coronaries since 1972. Querying the database to examine outcome is practical for research, but limited as a clinical tool. Thus, the data in the database are largely accessible through published datasets and prediction models. These prediction models have been reflected in a computer program that may run on a personal computer. In hospital outcomes are predicted by stepwise logistic regression and long term outcome (time-to-event) by Cox model analysis. Mean survival curves for the Cox model are needed as well as the coefficients. The data for survival curves is stored in an Access 2.0 database. The primary application is programmed in Visual Basic 3.0 and will run on any IBM PC or clone running DOS 5.0 or higher and Windows 3.0 or higher. The application uses a full graphical user interface. The in-hospital outcome data are expressed as percentages. Survival (time-to-event) data are expressed as survival curves. The entire application is mouse controlled and there is no typing at any time. The user starts the program from a windows icon and then a main menu offers a selection of models. The user clicks on a model. Then a specific screen for the model appears and the user selects values for the correlates of the endpoint. When done the user clicks for the data to be displayed. Movement within models and between models is straight forward, simple and fast. The current models after coronary surgery: in hospital death, Q wave myocardial infarction, stroke, death after reoperative surgery, long term survival, non-fatal events and survival after reoperative surgery. The current models after angioplasty: in hospital death, Q wave myocardial infarction, emergency surgery, restenosis, long term survival and long term cardiac events. Clinical cardiologists and cardiology fellows can learn to use the program essentially immediately, without training. Clinical understanding and proper use of the tool are the limitation as the computer application has been simplified as much as possible.