摘要：Background and aims: Variables in biology and medicine have a series of particularities when fitted to a curve, especially when their value is conditioned by time. Many curve fitting comparisons are based on the value of R squared, indicator which is not accurate in describing not-nested non-linear models. Material and method: We fitted a model of remaining quantity of viable cells affected by an immune-mediated destruction process in relation with disease duration on different types of curves, and we assessed the evidence ratio for each one with the proper indicator, the Akaike’s Information Criterion, then we investigated the relationship between this and value of R squared. Results: Linear curve fitting was not appropriate for our model. The values of R squared were disconcordant in relation to the proper indicator, the Aikake’s Information Criterion and Sum of Squares. Conclusion: Best fitting curve model for our example is the logarithmic one. In comparing not-nested, non-linear models in medicine Aikake’s Information Criterion should always be used in detriment of R squared.
关键词：curve fitting models; Akaike’s Information Criterion; R squared; Diabetes Mellitus
