In order to minimise the shortcomings of long-term predictions of bridge deterioration due to insufficient data, this paper presents an integrated method to build a reliable transition probability for predicting long-term performance of bridge elements. A selection process is developed in this method to automatically select the suitable prediction approach for a given situation of condition data. An Artificial Neural Network (ANN)-based Backward Prediction Model (BPM) is also employed for effective performance prediction when element inspection records are insufficient. A benchmark example is presented in this paper to demonstrate how the BPM-generated missing historical data together with available inspection data can be used as input for a state-based model in the integrated method. The outcome demonstrates that BPM-generated historical data, together with available data, can improve the reliability of transition probabilities, and in turn improve the reliability of long-term predictions.