Research model determined to be ineffective for capturing machine-learning appropriate data

Early detection of ongoing hemorrhage (OH) before onset of hemorrhagic shock is a universally acknowledged great unmet need, and particularly important after traumatic injury. Delays in the detection of OH are associated with a "failure to rescue" and a dramatic deterioration in prognosis once the onset of clinically frank shock has occurred. An early alert to the presence of OH would save countless lives.

This is a single site study, enrolling 48 patients undergoing liver resection in a "no significant risk" prospective clinical trial to: 1) further identify a minimal subset of noninvasive measurement technologies necessary for the desired diagnostic performance, 2) validate the performance of our Phase I algorithm, and 3) re-train the algorithm to a Phase II human iteration.

The main outcome variables are non-invasive measurements that will be used for machine learning, not real-time patient management. The data generated will be used later for discovery and validation in traditional and innovative machine learning.

Hemorrhagic shock remains a leading cause of death on the battlefield as well in civilian communities. Early detection of ongoing hemorrhage before progression to frank shock would allow early intervention. It is widely appreciated that the classic medical vital signs perform poorly until late in the progression to shock after traumatic injury. Currently available techniques, including intermittent vital sign monitoring, laboratory analysis, and single measurement devices have poor performance before clinically obvious physiologic distress.

The overall goal of this project is to develop a multi-technology noninvasive system for early detection of ongoing hemorrhage. The underlying hypothesis is that deep learning developed algorithms obtaining diagnostic signals from multiple sources will outperform single technology solutions.

While the promise of innovative noninvasive testing has received wide attention, development of effective bedside technologies has thus far been limited and their performance disappointing. In 2014, Kim et al stated that "The results from this meta-analysis found that inaccuracy and imprecision of continuous noninvasive arterial pressure monitoring devices are larger than what was defined as acceptable" and noninvasive blood pressure measurement is among the most fully developed of these technologies. The failure of noninvasive technologies in the detection or diagnosis of complex disease states has been essentially complete. The investigators believe that this failure reflects the limitations of uniplex systems (a single sensor in a single-location) and patient-to-patient variation in physiologic response. Uniplex systems sacrifice the entire diagnostic signal in anatomic-temporal patterns, which likely has significant discriminant power.

To date, technological innovation in early detection of ongoing hemorrhage has been of two broad categories: 1) a search to discover a single new measurement of tissue or organ status or 2) application of more sophisticated mathematical techniques based on machine learning and signal processing.

The investigators propose to develop a system that combines state-of-the-art noninvasive sensing technologies and advanced multivariable statistical algorithms. This system will be developed from its inception to be inexpensive and easily applied, even in austere settings.

To avoid the unnecessary use of blood products, hepatectomies are performed with low central venous pressure (CVP). This is accomplished through restrictive use of intravenous fluids and at times medications to lower the central venous pressure. Low central venous pressure during hepatectomy is an excellent model for development of technologies such as ours and has not been previously used for this purpose.

During each procedure, the investigators will obtain a full ensemble of noninvasive optical, electromagnetic and impedance physiological signals during the LCVPLR procedure. The work proposed herein will evaluate these technologies during standard low central venous pressure liver resections (LCVPLR). These data will be utilized for further machine learning-based algorithm development. The proposed study will be low risk since the measured data will not be available to the clinicians.

Specific Aims:

1. Evaluate the performance of existing non-invasive sensing technologies and multivariable algorithms in LCVPLR.
2. Obtain human model training and validation data sets during LCVPLR for further refinement of the algorithms.

Power and Sample Size: The investigators anticipate acquiring data from every enrolled subject. The data obtained before onset of parenchymal transection will be utilized as the "no hemorrhage" control. Power and sample size calculations indicate that a sample size of 48 subjects should be sufficient to: 1) further identify the minimal subset of noninvasive measurement technologies necessary for the desired diagnostic performance, 2) validate the existing algorithms, and 3) initially train a human clinical iteration of the algorithms, with a sufficient degree of accuracy (p < 0.05 for ROC-AUC).

As a minimal risk study, there will be no change from standard of care for patients undergoing surgery. The surgical procedures and pharmacotherapies will proceed as per standard clinical management. Enrolled patients will undergo standard preoperative, anesthetic, and postoperative physiological monitoring.

Inclusion Criteria:

1. Adults 18 years or older
2. Patients undergoing liver resection.
3. Ability to give informed consent.

Exclusion Criteria:

1. Pre-existing systemic illness, likely to alter systemic cardiovascular response to hemorrhage. Including congestive heart failure, and a paced cardiac rhythm.
2. Pregnant
3. Prisoner status

Patients undergoing liver resection