Statistical Analysis of Data Fom Infertility Patients: How to Explicitly Consider the Decline in Fertility Associated With Age

Abstract

The decline in fertility with age is well documented. The objective of this study was to evaluate multiple regression as a strategy to address changes in fertility associated with age. Retrospective chart review of 2573 patients down regulated with GnRH agonists. This group includes a broad spectrum of infertility patients with ages 22-44. The data set used for the preparation of 2004 ASRM Abstract P-293 (Beltsos et al., Fert Steril 2004; 82, Suppl 2: S239-240) was re-examined using multiple regression. Previously, two-way ANOVA and chi square tests were used to identify changes resulting from treatment with recombinant FSH (r-hFSH) alone versus a combination protocol consisting of hMG plus r-hFSH, in patients classified as younger (< 35 years) versus older (≥ 35 years). That is, the effect of age was noted by dividing patients into arbitrary age groups. Herein, the data were re-analyzed by multiple linear and logistic regression. A p < 0.05 was considered statistically significant. Two different types of responses were taken as prototypic. Oocytes Retrieved was taken as an example of continuous data while Clinical Pregnancy Rates were taken as an example of a categorical variable. Age was treated as an independent variable and was graphed on the abscissa. Patient responses were treated as dependent variables and were graphed on the ordinate. All responses examined were adequately described by equations that included an intercept and slope for AGE over the 20 to 44 age range. Multiple regression models included AGE and ADDITION OF hMG (hMG) as main effects plus an AGE * hMG interaction term. When the AGE * hMG interaction was not significant, the model was re-calculated without the interaction. For multiple linear regression, a non-significant interaction means that the data can be appropriately described as two parallel regression lines, one for r-hFSH+hMG and one for r-hFSH alone. The difference in position of the regression lines represents the effect of adding hMG. Figure 1 illustrates results obtained for the continuous variable Oocytes Retrieved. Treatment of patients with r-hFSH alone yielded an additional 1.4 oocytes retrieved over the entire 20 to 44 age range. Analogous results were obtained for the categorical variable Clinical Pregnancy Rate analyzed by multiple logistic regression. Irrespective of the patient’s age, r-hFSH protocols resulted in about 7.5% higher pregnancy rates compared to r-hFSH+hMG protocols. Multiple regression analysis appears to be superior to the two-way ANOVA and Chi square tests used previously, particularly if the data include a broad range of patient ages. Multiple linear and logistic regression for the analysis of continuous and categorical variables, respectively, can be used to examine the effects of factors that affect treatment outcome while explicitly addressing known changes in fertility associated with age. Dealing with age by multiple regression allows other factors affecting patient responses to be characterized and quantified in a statistically robust manner.