Abstract
Purpose:
There is growing interest in the role of cerebrospinal fluid pressure (CSFP) as a contributing factor in the pathogenesis of glaucoma. Currently, the only known way to measure CSFP is by lumbar puncture, an invasive procedure. Recently, a method to estimate CSFP using a regression formula was proposed by Jonas et al, PLoS One, 2014. We compared this formula with one of our own design using a large medical record dataset.
Methods:
The regression formula proposed by Jonas et al, derived from a Chinese dataset in Beijing (CSFP[mm Hg] = 0.44 x Body Mass Index[kg/m2] + 0.16 x Diastolic Blood Pressure[mm Hg] - 0.18 x Age[Years] - 1.91) was tested on a Mayo Clinic database containing the medical records of patients having undergone lumbar puncture at the Mayo Clinic (Rochester, MN) between 1996 and 2010 (n = 4378). Half of the patients were selected randomly to comprise a training sample and the remaining patients were used for validation. Using the training sample, a new general linear model was derived with similar physiologic parameters to those utilized by Jonas et al. and fit to the validation sample to test CSFP prediction. Intraclass correlation (ICC) was used to assess predicted and actual CSFP in the validation data set.
Results:
The Beijing study’s ICC between training and validation group was 0.71. The Beijing regression equation poorly predicted CSFP in the Mayo dataset (ICC=0.14 [0.11-0.17]). The regression formula obtained from the Mayo training set was: CSFP[mm Hg] = 9.620 + 0.080 x Body Mass Index[kg/m2] - 0.042 x Age - 0.926 x Sex[F] + 0.0262 x Diastolic blood pressure[mm Hg]. The ICC between Mayo predicted and actual CSFP in the validation sample was 0.28 (95% CI=0.24, 0.32).
Conclusions:
The equation derived for predicting CSFP from a prospective study in Beijing, China fared poorly against a large, retrospective dataset from the Mayo Clinic. The Mayo regression formula performed better, but still failed to accurately predict CSFP. The possible differences may be due to the retrospective nature of the Mayo Clinic dataset, differences in the populations studied, and differences in LP technique. We conclude that caution should be exercised in using estimated CSFP derived from clinical data.