Techniques are provided for predicting the onset of a heart condition within a patient based on impedance measurements. Briefly, overloads in fluid levels in the thorax and in ventricular myocardial mass within the patient are detected based on impedance signals sensed using implanted electrodes. The onset of certain heart conditions is then predicted based on the overloads. For example, pulmonary edema arising due to diastolic heart failure is predicted based on the detection of on-going overloads in both fluid levels and ventricular mass. Ventricular hypertrophy is detected based on an on-going ventricular mass overload without a sustained fluid overload. Various other heart conditions may also be predicted based on specific combinations of recent or on-going overloads. Evoked response is exploited to corroborate the predictions. Appropriate warning signals are generated and preemptive therapy is initiated.