BHI® developed a top 10 list of significant factors that contribute to successful predictive modeling based on our deep experience working with really big data.
Answering the “so what” question before you start
No matter how interesting your model’s output might be, if the analysis isn’t actionable, it won’t get used.
Using massive databases
Massive databases ensure that large enough cohorts, with multiple diseases and social factors, are represented.
Exploring available data
Exploring available data to understand strengths and weaknesses, such as data bias
Being open to new methods
Being willing to learn new data science methods for making predictions that result in innovative solutions
Negotiating with other experts
Negotiating between what’s desirable and what’s possible among data science, clinical, and product experts
Securing access to powerful computers
Providing a computing environment that can quickly and seamlessly process and analyze a large data set
Possessing patient knowledge
Having an accurate understanding of what actions are feasible and appropriate for patients
Paying attention to detail
Taking time to sufficiently plan, design, develop, and validate predictive models
Using correct metrics
Using correct metrics to judge predictive performance so that better decisions can be made
Choosing the right data science platform
Establishing a data science platform that can easily integrate a finished model into usable software products