Towards precision critical care management of blood pressure in hemorrhagic stroke patients using dynamic linear models

Figures

• 

  Fig 1. Systolic BP and concentrations of two BP meds over an ICU course. BP data and medication admins taken from an ICH patient in an ICU dataset from UPMC. Systolic BP can be measured using both non-invasive methods and an arterial line. Drug concentrations are estimated from the medication administration record based on drug pharmacokinetics data.

• 

  Fig 2. Spontaneous systolic BP for UPMC intracerebral hemorrhage compared to Britton et al. results. Mean SBP for UPMC ICH patients compared with Britton et al.’s results from stroke patients with and without existing hypertension. Error bars for UPMC data denote 1 standard deviation. For the last data point, we provided SBP on day 8, while Britton et al. only reported SBP data for day of discharge.

• 

  Fig 3. AIM-BP framework. The AIM-BP framework consists of two parts: The model specification (a structured dynamic linear model) and the parameter estimator (a MCMC-based sampler).

• 

  Fig 4. Sample simulated BP time courses for each clinical scenario. Each clinical scenario presents with a BP of around 220/110 mmHg, which is managed down to around 140 mmHg systolic using medication. Upon visual inspection, it is difficult to identify how each scenario’s patient responds differently to the medications administered to them. AIM-BP can identify the different effectiveness of medications in each scenario.

• 

  Fig 5. Experimental design. The accuracy and consistency of the AIM-BP parameter estimator is compared against Non-Linear Least Squares on a set of clinical scenarios.

• 

  Fig 6. AIM-BP vs NLLS for Clinical Scenario 1. Box and whisker plots of estimated parameters using non-linear least squares (NLLS) and AIM-BP for Clinical Scenario 1. Significantly lower variances of AIM-BP estimated parameters compared to NLLS are noted with asterisks and p-values.

• 

  Fig 7. AIM-BP vs NLLS for Clinical Scenario 2. Box and whisker plots of estimated parameters using non-linear least squares (NLLS) and AIM-BP for Clinical Scenario 2. Significantly lower variances of AIM-BP estimated parameters compared to NLLS are noted with asterisks and p-values.

• 

  Fig 8. AIM-BP vs NLLS for Clinical Scenario 3. Box and whisker plots of estimated parameters using non-linear least squares (NLLS) and AIM-BP for Clinical Scenario 3. Significantly lower variances of AIM-BP estimated parameters compared to NLLS are noted with asterisks and p-values.

• 

  Fig 9. AIM-BP vs NLLS for Clinical Scenario 4. Box and whisker plots of estimated parameters using non-linear least squares (NLLS) and AIM-BP for Clinical Scenario 4. Significantly lower variances of AIM-BP estimated parameters compared to NLLS are noted with asterisks and p-values.

• 

  Fig 10. AIM-BP homeostasis perturbation model SBP compared to actual SBP and drug concentrations. Example calculated homeostasis perturbation model SBP using estimated AIM-BP parameters compared to actual SBP and drug concentrations. Dot-dashed red lines represent 2 standard deviations of Q from the homeostasis perturbation model SBP.