Implications for bias in estimated predictive performance when using different approaches to handle missing data across validation and implementation of a clinical prediction model: examples from cardio-thoracic surgery and external validation of two clinical prediction models: RESECT-90 and Thoracoscore.