Early prognosis of heart diseases or related issues can save lives in an enormous way. Heart Disease (HD) is one of the leading reasons to high death rate all over the globe. Learning machines are playing a vital role in the non-invasive, less expensive and reliable prognosis of HD. Some essential biomarkers for HD prognosis using learning machines are blood sugar, heart rate, blood pressure, etc. Deep Learning (DL) models are found to be more accurate for HD detection in comparison to Machine Learning (ML) models. Further, Deep Learning Models work well with sufficiently huge amount of dataset. This study contributes a novel two-phased deep network comprising a Generative Adversarial Network (GAN) for data generation and a LSTM based deep network for HD prognosis. The proposed model is benchmarked against few selected HD prognosis models from the literature. The dataset from University of California (Irvine) naming Cleveland is used for experimentation. The statistical investigation reveals that the proposed two-phased HD prognosis model is effective with accuracy and f-measure of 99.6% and 99.63%, respectively.