Machine Learning and 3D Image-based Modeling for Body Weight Estimation.

Drug dosing errors can have a catastrophic effect in acutely ill patients such as stroke patients needing thrombolytic therapy or patients requiring urgent sedation. In an acutely ill patient, inaccurate weight estimates are a significant cause of dosing errors, and weight estimates that deviate by \>10% from actual weight could make treatment itself life threatening. Inaccurate weight estimates lead to inaccurate drug doses, which can result in potentially fatal treatment failure (from subtherapeutic doses) or potentially fatal adverse events (from supratherapeutic doses). Nearly 75% of treatment failures in obese patients may be related to errors in weight estimation. When clinical care is time-sensitive, it may be impossible to obtain a measured weight in \>50% of patients. In these circumstances, a rapid, accurate method for estimating weight is critical. One recent innovation is the use of a low-cost 3D camera system to estimate weight. The 3D camera device (e.g., Intel RealSense D415) is used to obtain a point cloud map of the patient's body, from which a weight estimate can be estimated based on algorithms derived using convoluted neural network analysis. Initial studies have been extremely promising in terms of the accuracy achievable by this system in estimating Total Body Weight (TBW).

The primary aim of this study is to measure the accuracy of weight estimations by the 3D camera system in acutely ill or injured ED patients and compare this accuracy against that of standard care. The researchers will compare the performance and downstream effects of weight estimation using the 3D camera system against standard care in a randomized controlled trial of acutely ill or injured adults presenting to the ED.

The key hypothesis is that the 3D camera system will provide real-time estimates of TBW, IBW and LBW in an emergency setting and will exceed the accuracy of existing methods of weight estimation.

Supporting non-clinical trial studies will establish the accuracy of the 3D camera system in laboratory conditions, and in simulated medical emergencies. However, its performance, and its impact on downstream drug dosing accuracy, needs to be established during emergency care in a real clinical setting. This study will provide an essential perspective about the accuracy and functioning of the 3D camera system as well as real-world weight estimation during emergency care. It will also describe the ability to measure weight using in-bed scales and to obtain weight estimations from patients themselves and family members in ED patients. The secondary objective, to determine the accuracy of drug doses in each arm of the study, will provide critical information on the need for alternative weight scalars in obese and morbidly obese patients presenting to the ED. The study will establish the need for standards and policies to guide dose scaling in obese patients in the ED. Information on the actual usage of drugs that should be scaled to TBW and those that should be scaled to LBW will provide useful real-world insight into the magnitude of the problem in the threat to patient safety by using a "one size fits all" approach to drug dose calculations for all patients, irrespective of weight status.

Acutely ill patients presenting to the ED of a large regional hospital, and who require weight-based drug therapy, will be enrolled in the study. They will be randomised to either receive a weight estimation using a 3D camera system (which will provide estimates of TBW, IBW and LBW), or to receive standard care. All other interventions and medical care will be standard care.

These patients will be followed for the first 72 hours of their hospital stay. The accuracy of the weight estimates will be compared between the groups, as will the drug dose accuracy, and any adverse events related to drug therapy.

• Accuracy of in-bed scales [Time frame: Immediate]