Repeatability and variance analysis on multiple computer-assisted (IVOS*) sperm morphology readings
The repeatability of the Hamilton Thorne Research IVOS (version 10) semen analyser (dimension specific software, version 3) in the evaluation of sperm morphology according to strict criteria was investigated in this study. The repeat measures investigated were cell-cell (300 cells, 3 x each), intraslide (20 slides, 3 x each) and interslide (30 samples, 3 slides each), and their normal sperm morphology outcomes were recorded. Semen samples with varying normal sperm morphology percentages were obtained and sperm morphology slides prepared. The slides were stained with Diff-Quik stain. Agreements between evaluations were determined using the κ statistic and average coefficients of variation. The predictive probability for an abnormal cell given a prior abnormal cell outcome was 91%, and 89% for a similar prediction of a normal cell. The predictive probabilities for an abnormal or a normal cell given two prior abnormal or two prior normal cell outcomes were 95% and 94%, respectively. No significant bias was obtained between the repeat probabilities for normal and abnormal sperm cells. The average coefficients of variation for the intraslide trial were 9.73% and 8.30% when 100 and 200 sperm cells were evaluated, respectively. The average coefficient of variation for the interslide trial was 15.39%. The technical importance of good sample and slide preparation technique has once again been highlighted by this study. A uniform (spatial homogeneity), high concentration (5-10 cells per computer screen) smear must be made and the cells stained with optimal intensity (maximum contrast). In a trial in which 2000 cells were evaluated, 19 objects (0.95%) were identified as spermatozoa, but were debris. The automated semen analysing system (IVOS) used in this study was shown to maintain a level of repeatability, precision and accuracy acceptable for the application of the system in a routine semen analysis situation.