| Sumario: | Background: Previous studies showed a correlation between intra-abdominal-pressure (IAP) and body anthropomorphic data like sagittal-abdominal-diameter and body-mass--index (BMI) [1, 2]. Objectives: The aim of this study is to examine possible relations between other body parameters and baseline IAP in critically ill patients. Furthermore, this study will also compare gastric versus bladder pressure measurements. Methods: Prospective study in 96 mechanically-ventilated patients equipped with a Foley bladder catheter connected to a FoleyManometer (Holtech Medical, Charlottenlund, Denmark) to measure intrabladder pressure (IBP) and a CiMON balloon-tipped nasogastric probe that records endexpiratory (IAPee), endinspiratory (IAPei) and mean IAP (IAP) (Pulsion Medical Systems, Maquet Getinge Group, Feldkrichen, Germany). Intra-abdominal hypertension (IAH) is defined as an IBP or IAPee above 12 mm Hg and ΔIAP is defined as IAPee-IAPei. Comparison of bladder (IBP) and gastric (IGP) pressure measurements was done with Pearson correlation and Bland and Altman analysis. The following body anthropomorphic parameters were measured: distances (ear-xiphoid, ear-nose, xiphoid-pubis, and ribcage-crista), diameters (rib cage, umbilical, waist, and hip), circumference (rib cage, abdominal, waist, and hip), height (patient, rib cage, hip, and sagittal abdominal diameter). Furthermore, a simulation model was developed with Datastories (www. datastories.com) in order to identify independent anthropomorphic and other parameters (out of a total of 28) able to predict IBP. We had to create and challenge 66,537 predictive models to deeply learn which metrics are necessary and sufficient to predict IBP. A half of the computational effort was spent on meticulous cross-validations to make sure to avoid over-fitting and maximizing the predictive power of models. At the end, we were able to build a final ensemble of 100 models with a minimal number of metrics. With respect to predicting the IBP, the health of our data was satisfactory. DataStories focuses on finding reliable relationships between numeric metrics and IBP. Therefore, we had to look at 28 metrics remaining after omitting 5. We first looked at how our metrics impact the IBP individually and performed a standard correlation analysis as well as a more involved analysis of the mutual information content between IBP and all other data inputs individually. Because the data studied only had 28 parameters on top of the IBP we also computed all individual pairwise relationships (correlation and mutual information) among the metrics to see how parameters are connected to each other (Fig. 1). Based on initial results we could conclude that 1 out of 28 inputs could be removed from the consideration whatsoever, because it did not have even slight independent relationship to IBP. Results: SAPS-II was 55.4 ± 12.9; APACHE-II 26.4 ± 9.6, SOFA 11.3 ± 5.2; age 57.5 ± 13.9; height 174 ± 9 cm; weight 85 ± 20; BMI 27.9 ± 7. The patients with IAH (n = 55) had higher BMI (29 ± 8 vs 26 ± 4, P = 0.02). The ΔIAP was significantly higher in IAH: 5 ± 1 vs 3 ± 1 mm Hg (P < 0.0001). We found a positive correlation between IAP and ΔIAP, suggesting a lower abdominal wall compliance (Cab) the higher the IAP: ΔIAP = 0.3 × IAP + 0.1 (P < 0.001, R2 = 0.579). The following body parameters were significantly higher in patients with IAH: IAPmean (15 ± 3 vs 9 ± 2 mm Hg), umbilical diameter (43 ± 7 vs 40 ± 5 cm, P = 0.02), abdominal perimeter (121 ± 17 vs 108 ± 11 cm, P < 0.0001), waist circumference (107 ± 14 vs 100 ± 11 cm, p = 0.01), the convex xiphoid to pubis distance (40 ± 7 vs 35 ± 5 cm, P < 0.0001), rib cage height (24 ± 3 vs 21 ± 4, P = 0.0001), sagittal abdominal diameter (28 ± 4 vs 21 ± 4 cm, P < 0.0001). Patients with IAH had higher alveolar plateau pressures (29 ± 5 vs 25 ± 5 cm H2O, P = 0.0002) and higher PEEP (10 ± 3 vs 8 ± 3 cm H2O, P = 0.04). Patients with IAH had lower abdominal compliance, defined as ΔTV/ ΔIAP (137 ± 55 vs 222 ± 85, P < 0.0001). Significant differences were observed between men and women. There was a significant Pearson correlation between IBP and IGP (IBP = 1.04 × IGP + 1.1 mm Hg, R2 = 0.91, P < 0.0001). Bland and Altman analysis comparing IGP and IBP at endexpiration showed a mean bias of 1.6 ± 1.1 mm Hg. The limits of agreement were small from -0.7 to 3.9 mm Hg resulting in a percentage error of 26%. The final ensemble of parameters after data analysis able to predict IBP has the following characteristics: average cross-validation prediction accuracy of 74.5% with the use of 3 metrics (out of the initial 28 that were entered into the model). From this preliminary analysis, we could conclude that 27 inputs are individually related to IBP, but many of them are correlated to each other (Fig. 2). Therefore, further datamining via Datastories was able to identify a minimal set of 3 metrics that matter in order to predict IBP (Fig. 3). After deeply learning our prediction problem by creating and challenging 66537 models, we have discovered that 3 metrics are sufficient to predict IBP at Pearson correlation of 0.745 (R2 = 0.56). These driver metrics have various influence on the IBP and have to be used together to make robust predictions (Fig. 4). The drivers are the difference between the convex and horizontal xiphoid to pubis distance (importance: 37%), the sagittal abdominal diameter (SAD, importance: 4%), and the abdominal compliance (Cab, importance: 58%), all together their importance sum up to 100%. Conclusions: Patients with IAH have increased waist and abdominal perimeter, convex xiphoid-to-pubis distance, rib cage height, and sagittal abdominal diameter. Female patients have significantly different body measurements. High IAP is related to ΔIAP and low Cab. Deep learning identified 3 independent factors able to predict IBP with 74.4% accuracy: SAD, Cab and the difference between the convex and horizontal xiphoid to pubis distance. Body anthropomorphy plays a role in the abdominal wall compliance and the way the patients IAP behaves in relation to increased intra-abdominal volume. In our patient sample, we found a good correlation between IGP and IBP when measured at endexpiration in supine position.
|
|---|