Input-output learning model for quantitative analyses for biomedical researchers

29 July 2013

Ask one of the multiple old school statisticians, and he will tell you that unless a clinical researcher knows absolutely everything that there is to be known about a test or model you should not even touch it. Instead, let them do everything and then provide the researcher with an answer that should be taken as truth. Kind of like an oracle.

The problem with this concept is that data analysis is now part of virtually any clinical study, and keeping one side blinded to what might be happening quantitatively is, let's say, not a good idea. While the attempt might be to simply protect some turf or genuinely avoid misinterpretation, the adverse consequences are that both sides will be blind sighted: the clinical researcher by not understanding the analysis, but also the statistician by not being able to fully obtain information from the clinical researcher.

But how do you increase the communication in this trading zone? While there is no single answer, one attractive educational model is the idea of treating tests and models as an input-output mechanism. In other words, start by ignoring the internal mathematical mechanics of the test or model and treat it as a device that requires a certain type of questions (variables), and that will spit out results that can be interpreted in a certain way. Further, display the output as a series of tables and graphics that clinical researchers are likely to find in their research papers, so that they can immediately connect that to the study objectives.

For example, if we treat a logistic regression model as an input-output device, then it will require a dichotomous (yes/no) variable as input to be predicted, a group of variables or any kind as predictors. The output will have a series odds ratios that can be interpreted as the risk increase in the occurrence of the outcome associated with a given predictor. Present a table, and then show exactly what an odds ratio is, what the confidence interval is and what it means.

Now, does the input-output model provides for a comprehensive learning experience? Absolutely not, but that's not what we want when getting a clinical researchers started on a quantitative topic. Instead, what they need is a sense of mastery on a simplified task, something that will give them the confidence to start a discussion with the quantitative experts and, then, the interest to start learning beyond that initial whole task.

by Ricardo Pietrobon

About Me

My name is Ricardo Pietrobon and I am interested in big data and situated cognition applied to immersive distance education.