Objective: Machine learning methods are pervasive in fields such as image and speech recognition, advertising, and fraud detection. But their adoption in health economics and outcomes research (HEOR) has been slower, and potential applications of machine learning as an addition to the traditional statistical approaches have not yet been fully recognized in the field. This study aims to develop a simple set of if-else rules (a simple decision tree) to help HEOR professionals identify research questions where machine learning can help improve outcomes obtained from traditional statistical analysis. Methods: We conducted a search on PubMed to retrieve studies that compare the predictive performance of machine learning methods against statistical methods in the last ten years, which yielded 688 abstracts for review. We summarized application areas where machine learning methods performance was worse, comparable or better compared to traditional statistical methods. We then summarized the arguments on the merits of both methods in health care based on online discussion forums and blogs from authors whom we could identify as either machine learning practitioners or statisticians. Using the results of our analysis and review, we developed a simple decision-tree like algorithm as a general guide to classify research questions as falling into the application domain of machine learning or statistical analysis and give examples from the literature. Results: Machine learning methods generally outperform traditional statistical methods for prediction tasks, although the performance improvement varies across studies. Increased complexity of machine learning methods should be weighed against the gain in predictive performance. Conclusion: Following relatively simple rules of thumb can point at the right method to answer research questions. Recognizing the appropriate application areas for machine learning can increase the speed of adoption of these techniques in HEOR.