How to avoid errors when computing reference interval tables and graphs using regression equations for cross-sectional studies of fetal biometry.

Abstract

While visualization of fetal anatomy is one of the primary purposes of diagnostic ultrasound, the application of statistical analysis to biological measurements of fetal structures such as the head, abdomen, long bones, and heart, requires equations that compute changes in these measurements over time. One of the challenges for the investigator who desires to compute reference intervals for a fetal measurement is selecting the best equation for the measurements of interest. Fractional polynomial regression equations have been used to describe changes in fetal measurements over time and consists of 44 equations for the investigator to select. Since it would be laborious to analyze each equation to identify the best fit of the data, software programs have been developed for such analysis. However, depending upon the software program, potential errors can occur when translating the coefficients of a complex equation to a format that can be used by the clinician to create reference interval tables and graphs. This study reviews the requirements for statistical output that allows the investigator to generate reference interval tables and graphs using Excel. A step-by-step sequence is explained that enables the investigator to minimize the possibility of errors when reporting complex equations and their associated graphs that result from fractional polynomial regression analysis. Using the techniques described in this opinion paper should decrease the potential for reporting erroneous equations used to compute reference interval tables and graphs. This article is protected by copyright. All rights reserved.