Is delayed relaxation independent from global diastolic dysfunction?
This is a comprehensive dataset of 60 children with Coarct, AS, KD with aneurysms or HTx (all high risk for LVDD) who had simultaneous cath and echo data acquired. The objective was to assess if delayed relaxation and global diastolic dysfunction were independent of each other. The secondary objective (not explored here) was to validate echo tools of DD against GS (Tau and LVEDP).
As you can see here, Tau has a moderate correlation with LVEDP (r = 0.576) which may refute our hypothesis. In addition, the linear model suggests that LVEDP could indeed be associated with Tau. However, is that the case?
Subgroup analysis
The most telling thing about these boxplots is that CoA has high variance as evidenced by all the outliers and large IQR. Also what is interesting is that the medians of tau are pretty similar. KD is the only group that has a low median LVEDP, but the IQR has significant overlap in LVEDP.
One of the hypotheses is:when there is DD, LVEDP should be high, and tau should be normal. The median LVEDP is 12.7 (>50% of pts had LVEDP >12). Let’s visualize what values of tau were associated with high lvedp (>12). If high LVEDP is associated with high tau, then that would refute the hypothesis.
Do patients with high Tau have a high cutoff?
CoA appears to have a clear distinction between hi/lo LVEDP and tau while the other ones not so much.
Is there a linear relationship between Tau and LVEDP by diagnostic subgroup?
Clearly, CoA has an obviously strong relationship, and might be even better explained with a logarithmic type equation. KD is clearly weak. However, AS and HTx appear to be pretty random as well with only a few points likely generating the strenght of correlation present in the coefficient.
Correlation coefficients: CoA 0.78 HTx 0.50 AS 0.40 KD -0.02
Next step: Model the data
| diagnostic_group | term | estimate | std_error | statistic…8 | p_value | lower_ci | upper_ci | r.squared | adj.r.squared |
|---|---|---|---|---|---|---|---|---|---|
| CoA | intercept | 2.029 | 2.358 | 0.861 | 0.401 | -2.925 | 6.983 | 0.6050828 | 0.5831429 |
| CoA | tau | 0.269 | 0.051 | 5.252 | 0.000 | 0.161 | 0.377 | 0.6050828 | 0.5831429 |
| AS | intercept | 7.879 | 5.925 | 1.330 | 0.213 | -5.322 | 21.081 | 0.1579380 | 0.0737318 |
| AS | tau | 0.190 | 0.139 | 1.370 | 0.201 | -0.119 | 0.500 | 0.1579380 | 0.0737318 |
| HTx | intercept | 0.263 | 7.037 | 0.037 | 0.971 | -15.657 | 16.182 | 0.2450668 | 0.1611853 |
| HTx | tau | 0.353 | 0.207 | 1.709 | 0.122 | -0.114 | 0.821 | 0.2450668 | 0.1611853 |
| Kawasaki | intercept | 11.098 | 7.286 | 1.523 | 0.150 | -4.529 | 26.726 | 0.0005418 | -0.0708481 |
| Kawasaki | tau | -0.018 | 0.205 | -0.087 | 0.932 | -0.458 | 0.422 | 0.0005418 | -0.0708481 |
With all the data together, the linear model says that beta coefficient of tau is 0.25, with a significant P-value, but RMSE is 4.5, and adj R2 is 0.32.
If you run a linear model by diagnostic subgroup, The ONLY significant coefficient here is CoA. AS, HTX, and KD all have insignificant coefficients! In addition, adjusted R2 of Coa is 58% while all others are <16%. This is highly suggestive that only CoA correlates. The physiology is a very interesting discussion to have. However, the caveat is that we may just not be powered to find significant differences in the other groups (n=12-16) while CoA has the highest n of 20.
A more unifying model would be a mixed effects model where I could account for the 4 different groups of tau by dx subgroups. Unfortunately don’t know how to do that, yet.