Morphology cohort
Factors affecting clinical pregnancy
One of the most common questions from both patients and physicians is simple on the surface, yet complex in practice: which embryo should be selected for transfer?
Answering this requires weighing several key factors—blastocyst expansion (grades 1–6), inner cell mass (ICM) quality (A–B-C), trophectoderm (TE) quality (A–B-C), and the day of development at freezing (Day 5, 6, or 7). Even these four variables alone generate 216 possible embryo profiles (6 × 3 × 3 × 3), making intuitive decision-making both challenging and inconsistent.
To move beyond subjective assessment and address this question rigorously, we applied a data-driven approach. Using binary logistic regression, we modeled clinical pregnancy as the outcome, incorporating maternal age, biopsy day, number of embryos transferred, and standardized morphology metrics (expansion, ICM, and TE derived from Gardner grading). In parallel, principal component analysis (PCA) was used to better understand the structure and relative contribution of these correlated features.
All predictors were standardized (z-scores), allowing direct comparison of effect sizes. Results are presented as odds ratios per one standard deviation, providing a clear interpretation of each variable’s impact.
When reviewing the results, use both the tables and charts together:
- Tables provide precise estimates and uncertainty
- Bar charts highlight the relative magnitude and direction of each effect
This combined approach helps translate a complex, multi-dimensional decision into something quantifiable, reproducible, and clinically actionable.
Feature importance for clinical pregnancy (logistic regression)
FET cycles (model)
4,596
Clinical pregnancies (IUP)
3,329
IUP rate
72.4%
ROC AUC (discrimination)
0.621
Most Important factors for embryo selection (Relative importance (|β| on z-scale)
After z-scoring every predictor, the fitted logistic model assigns each one a coefficient β on the log-odds scale. Because scales are comparable, the absolute value |β| is a simple index of how strongly that variable moves predicted IUP risk when it increases by one cohort standard deviation.
How to read the figure: vertical axis = |β| (larger → stronger association in this model). Bars are ordered by |β| (largest on the left). This ranks contributors within this regression; it does not by itself prove which factors are clinically most important.
Figure: magnitude of standardized logistic coefficients (|β|) for each predictor; same estimates as in the regression table below.
ROC curve (discrimination)
Each point corresponds to a threshold on the model's predicted probability of IUP (scores sorted from high to low). The blue curve is this logistic model; the dashed gray line is chance (no discrimination, AUC = 0.5). AUC = 0.621 is the area under the blue curve (trapezoidal rule, consistent with the rank-sum AUC reported above).
Square plot: both axes are 0–1. Curves above the diagonal indicate better-than-random ranking of IUP vs non-IUP by predicted probability.
Regression table
Predictors are z-scored before fitting. Odds ratios ("OR") are for a one standard-deviation increase in the predictor. Wald z and two-sided p-values are approximate.
| Predictor | β (log-odds) | SE | OR | 95% CI | z | p |
|---|---|---|---|---|---|---|
| Intercept | 0.9296 | 0.0333 | — | — | 27.96 | <0.0001 |
| Embryo age (years) | -0.0024 | 0.0330 | 0.998 | 0.935 – 1.064 | -0.07 | 0.942 |
| Biopsy day (day 5 or 6) | -0.2604 | 0.0321 | 0.771 | 0.724 – 0.821 | -8.10 | <0.0001 |
| Embryos for ET (count) | 0.0296 | 0.0334 | 1.030 | 0.965 – 1.100 | 0.89 | 0.375 |
| Expansion grade (1–6) | -0.0817 | 0.0331 | 0.922 | 0.864 – 0.983 | -2.47 | 0.014 |
| ICM grade (ordinal A/B/C) | 0.2175 | 0.0315 | 1.243 | 1.169 – 1.322 | 6.91 | <0.0001 |
| TE grade (ordinal A/B/C) | 0.1204 | 0.0332 | 1.128 | 1.057 – 1.204 | 3.62 | 0.0003 |
Odds ratios (per 1 SD)
Each bar is the odds ratio (OR) for a one standard-deviation increase in that predictor, holding the model structure fixed. OR = 1 means no association with the odds of IUP; OR > 1 means higher predicted odds of IUP per SD increase; OR < 1 means lower.
Reference line: the horizontal dashed line at OR = 1 marks “no effect” on the odds scale. Bars entirely above that line suggest a positive association with IUP on average in this cohort; interpret together with confidence intervals in the table.
Figure: odds ratios per 1 SD (same point estimates as OR column in the regression table); vertical scale fits the cohort’s OR spread.
How to read this
- Outcome: Clinical pregnancy vs non-pregnancy
- Predictors: numeric embryo age, biopsy day, embryos for ET, and Gardner expansion plus ordinal ICM/TE scores from Gardner letter grades.
- Standardization: each predictor is mean-centered and scaled by its standard deviation so coefficients are comparable in magnitude.
- AUC / ROC: the ROC plot shows sensitivity vs false positive rate for all classification thresholds; the numeric AUC summarizes how well predicted probabilities rank IUP vs non-IUP (1 = perfect, 0.5 = random).
Principal components analysis of features affecting clinical pregnancy
Features: embryo age, biopsy day (5/6/7), embryos for transfer, Gardner expansion, morphology quality (ordinal), and ICM / TE letter scores from Gardner grades. All features are z-scored; orthogonal factors are derived from the correlation structure (variance-maximizing, ordered). Outcome (IUP) is used only for coloring and correlation with factor scores. A written interpretation of variance, loadings, and outcome alignment appears below the tables.
What each factor represents (this cohort)
Names list the two features with largest absolute loadings on that factor after standardizing all inputs. They describe which variables move together—not a clinical diagnosis name.
- Morphology qualityICM
Morphology quality · ICM (This is the factor that explains the most variance in the cohort.) Standardized loadings for the strongest contributors (in order): Morphology quality (+0.59), ICM (+0.58), TE (+0.41). A positive sign means that variable tends to increase when scores on this factor increase.
- ExpansionBiopsy day
Expansion · Biopsy day (This factor is orthogonal to all previous factors.) Standardized loadings for the strongest contributors (in order): Expansion (-0.75), Biopsy day (-0.54), TE (-0.31). A positive sign means that variable tends to increase when scores on this factor increase.
- EmbryosAge
Embryos · Age (This factor is orthogonal to all previous factors.) Standardized loadings for the strongest contributors (in order): Embryos (+0.70), Age (-0.69), Expansion (+0.15). A positive sign means that variable tends to increase when scores on this factor increase.
- EmbryosAge
Embryos · Age (This factor is orthogonal to all previous factors.) Standardized loadings for the strongest contributors (in order): Embryos (+0.70), Age (+0.70), ICM (+0.10). A positive sign means that variable tends to increase when scores on this factor increase.
Sample & variance
4,596 FET records, 7 standardized features.
| Factor | Variance explained | Cumulative | r (factor, IUP) |
|---|---|---|---|
Morphology qualityICM | 32.8% | 32.8% | 0.183 |
ExpansionBiopsy day | 18.2% | 51.0% | 0.080 |
EmbryosAge | 15.1% | 66.1% | 0.006 |
EmbryosAge | 13.2% | 79.3% | 0.039 |
Variance explained (scree)
Each bar is the share of total variance across standardized features captured by that factor (ordered by importance). Steep drops mean later factors add little.
Cumulative variance
How much of the total variance is retained when including factors in order. With seven inputs, all factors together reach 100%.
Loadings heatmap
Each cell is the loading of one standardized feature on one factor (eigenvector element). Blue = negative, red = positive; stronger color = larger magnitude. Compare with the numeric table below.
Factor scores vs clinical pregnancy (IUP)
Point-biserial r between each factor score and IUP (1 = pregnancy, 0 = not). Bars show direction: to the right = higher scores associate with IUP on average.
Score distributions by outcome
Overlaid histograms for the full cohort: green = IUP, gray = not Pregnant. Same bins; taller bar = more cycles in that score range. Overlap shows that outcomes are not separated by a single linear factor.
Morphology qualityICM
n = 3,329 IUP · 1,267 not IUP (same bins; bar height ∝ count in bin)
ExpansionBiopsy day
n = 3,329 IUP · 1,267 not IUP (same bins; bar height ∝ count in bin)
Loadings (first two factors)
Each column is the contribution of that standardized feature to the factor (direction in feature space).
| Feature | Morphology qualityICM | ExpansionBiopsy day |
|---|---|---|
| Embryo age | -0.085 | -0.110 |
| Biopsy day | -0.362 | -0.538 |
| Embryos for ET | -0.078 | 0.084 |
| Gardner expansion | -0.100 | -0.751 |
| Morphology quality (ordinal) | 0.587 | -0.165 |
| ICM grade (score) | 0.575 | -0.087 |
| TE grade (score) | 0.412 | -0.306 |
Interpretation of findings
Data-driven narrative for this cohort
This analysis derived orthogonal variance factors from 4,596 FET records and 7 standardized inputs: embryo age, biopsy day, embryos transferred, Gardner expansion, morphology quality (ordinal), and ICM/TE letter scores. The method finds orthogonal factors in which the cohort varies the most (variance-ranked). Clinical pregnancy (IUP) was not used to build those factors—they summarize patterns among the predictors only. Associations with IUP are computed afterward (correlation of factor scores with outcome).
The factor summarized as Morphology quality · ICM accounts for about 32.8% of total variance across standardized features (largest share). The features with the largest absolute loadings on this factor are Morphology quality (ordinal), ICM grade (score), TE grade (score) (see loadings table for signs). A positive loading means that variable tends to increase when moving in the positive direction along this factor in this cohort, and vice versa. This factor is usually the dominant mix of age, morphology, and expansion because they co-vary in real cohorts.
The factor summarized as Expansion · Biopsy day is orthogonal to Morphology quality · ICM and explains about 18.2% of variance separately. It is driven most by Gardner expansion, Biopsy day, TE grade (score). It often captures another pattern (for example differences between biopsy day and morphology that Morphology quality · ICM does not already explain).
Together, the 4 factors in the table (each labeled by its short summary) explain 79.3% of total variance. With only seven input variables, at most seven dimensions exist; factors with smaller variance shares are often harder to interpret.
- Morphology quality · ICM vs clinical pregnancy. Point-biserial correlation r ≈ 0.183 (modest linear association). On average, cycles with a clinical pregnancy (IUP) have higher scores on Morphology quality · ICM than cycles without (linear summary only).
- Expansion · Biopsy day vs clinical pregnancy. r ≈ 0.080. Expansion · Biopsy day can align with outcome differently from Morphology quality · ICM because it summarizes a distinct part of predictor space.
- Scatter plot (Morphology quality · ICM vs Expansion · Biopsy day). Overlap between green (Pregnant) and gray (not Pregnant) points indicates that these two linear summaries do not separate outcomes cleanly—pregnancy still occurs across many regions of the plot. This analysis is not optimized for prediction.
- Caveats. Results are exploratory and cohort-specific. They do not prove causation. Confounding factors are not adjusted here. Use these findings to generate hypotheses, not to counsel individual patients from this page alone.
Scores: Morphology quality · ICM (horizontal) vs Expansion · Biopsy day (vertical)
Subsampled points for display (even spacing along the sorted row index). Green = clinical pregnancy (IUP).