Machine learning-based clustering to identify the combined effect of the DNA fragmentation index and conventional semen parameters on in vitro fertilization outcomes

Abstract

Background: Previous studies have demonstrated an association between male sperm quality and assisted reproduction outcomes, focusing on the effects of individual parameters and reaching controversial conclusions. The WHO 6th edition manual highlights a new semen assay, the sperm DNA fragmentation index, for use after routine semen examination. However, the combined effect of the sperm DNA fragmentation index (DFI) and routine semen parameters remains largely unknown. Methods: We assessed the combined effect of the sperm DFI and conventional semen parameters on single fresh conventional IVF outcomes for infertile couples from January 1, 2017, to December 31, 2020. IVF outcomes were obtained from the cohort database follow-up records of the Clinical Reproductive Medicine Management System of the Third Affiliated Hospital of Guangzhou Medical University. An unsupervised K-means clustering method was applied to classify participants into several coexposure pattern groups. A multivariate logistic regression model was used for statistical analysis. Results: A total of 549 live births among 1258 couples occurred during the follow-up period. A linear exposure–response relationship was observed among the sperm DFI, sperm motility, and IVF outcomes. In multivariable adjustment, increased sperm DFI values and decreased sperm motility and semen concentration levels were associated with reduced odds of favourable IVF outcomes. Four coexposure patterns were generated based on the sperm DFI and the studied semen parameters, as follows: Cluster 1 (low sperm DFI values and high sperm motility and semen concentration levels), Cluster 2 (low sperm DFI values and moderate sperm motility and semen concentration levels), Cluster 3 (low sperm DFI values and low sperm motility and semen concentration levels) and Cluster 4 (high sperm DFI values and low sperm motility and semen concentration levels). Compared with those in Cluster 1, participants in Cluster 3 and Cluster 4 had lower odds of a live birth outcome, with odds ratios (95% confidence intervals [CIs]) of 0.733 (0.537, 0.998) and 0.620 (0.394, 0.967), respectively. Conclusions: When combined with low sperm DFI values, there was no significant difference between high or moderate sperm concentration and motility levels, and both were associated with favourable IVF outcomes. Low sperm parameter levels, even when DFI values remain low, may still lead to poor IVF outcomes. Participants with high sperm DFI values and low sperm motility and semen concentration levels had the worst outcomes. Our findings offer a novel perspective for exploring the joint effects of sperm DFI and routine semen parameter values.