The development of clinical prediction rules

Clinical decision instruments (CDIs), also known as "clinical decision rules" or "clinical prediction rules," are constellations of history or physical examination findings, and sometimes laboratory test results, that can be used as valid and reliable probability estimates for a disease or outcome. The methods to develop a meaningful CDI include careful contemplation about the outcome variable and the health care setting because not every clinical condition requires a CDI. Once the need for a CDI has been identified, multiple steps are necessary to maximize the reliability, accuracy, and usefulness of the instrument created. The first step is to derive the CDI via a search for variables that will best predict the outcome of interest. The next step is to validate the CDI by prospectively assessing how it performs, preferably in populations different to that in which it was derived. If the CDI is validated the final step is to assess for clinical impact. The predictor variables used in a CDI should be explicitly defined and easy to apply. Loosely defined terms or variables that may be interpreted differently by different clinicians (i.e., asking "Is the patient obese?" without any formal definition of obese) may result in poor performance when trying to validate the CDI. Recursive partitioning is generally superior to other types of modeling in the statistical derivation of CDIs. Clinical expertise and intuition is often an important complement to CDIs.