Sunday, February 17, 2008

Pediatric M-Mode Z-Scores

(Note: the calculator is now HERE )

I finished off what I thought were the more useful z-score calculations based on data from the article "Normal values of M mode echocardiographic measurements of more than 2000 healthy infants and children in central Europe" (link). Included in this calculator are z-scores and normal ranges for:

  • right ventricle (RVDD)
  • ventricular septum (IVS)
  • left ventricle end diastolic dimension (LVEDD)
  • left ventricle end systolic dimension (LVESD)
  • posterior wall (LVPW)
  • left atrial diameter (LAD)

I excluded calculations of the RV free wall, wall thickness in systole, and the arterial diameters- in part because I am not routinely reporting these measurements, but also because I question the application (PA diameter by m-mode??). Also, I hesitate to include the calculation of the left atrial z-score because reporting left atrial volumes is so much more descriptive, but alas I am not prepared to calculate LA volume z-scores for pediatrics...

I tried to use the published regression equations, both for the sake of simplifying the calculation and for the "continuously variable" effect, but ultimately abandoned this approach for several reasons:

  • The regression often did not yield the same result as the tabular data. For instance, using the formula for LVEDD for a child with a BSA of 0.5 m2 predicts a diameter of 28.19 mm; using the published table, the value is shown as 29.0 mm. The IVS predicted by the equation for patients with BSA's of 0.25 m2 and 2.0 m2  is 3.48 mm and 7.5 mm respectively; the same data from the tables is 3.8 mm and 9.3 mm.
  • I had to build a "lookup" routine for the standard deviations. As the authors did not publish prediction equations for the standard deviations, the values had to be hashed from a table anyhow. Adding another lookup table for the mean values was not much more work.
  • Infants were grouped by weight rather than BSA. Because, according to the authors "body surface area changes only minimally." The authors do not explore this topic any further though it begs the question of how the correlations were affected, i.e., what was the correlation coefficient when the infant's BSA was included in the analysis, and how much was it improved by removing them from the analysis? In part, the decision to break the infants out of the BSA relationship probably stems from a lack of understanding of the underlying relationship, as was recently and elegantly described by Sluysmans and Colan. Anyhow, because they grouped infants in this manner, z-scores could not be calculated using their prediction equations unless I mixed and matched the techniques (predict the mean using the BSA-based equation, and use tables to lookup the weight-based standard deviations).

In summation, this calculator simply addresses the published tables and performs lookup routines for the mean and standard deviation based on the appropriate index: weight for infants 2-4kg, BSA for subjects with BSA >0.25.

Update 6/2008: I re-worked the calculator to be quicker and prettier...

M-Mode Z-Score Calculator