Progress Report, May 27

Agent Behavior

The general conceptual model is as follows:

My main questions at this point is how to best model the care process.

Setup

• Initialize 20 beds, selectable amount of caregivers, initialize patients (apache and severity)
• Sickest patients get assigned individual caregivers
• Remaining caregivers are assigned to less acute patients. 2 patients per caregiver.

To go -

• I flow in ICU. I am either admitted or I overflow to another unit (how might I set up arrival distribution)?
• I am cared for and have a probability of dying based on my initial apache. (What is the most reasonable method of abstracting the care process?)
• If my apache decreases to 0, I am discharged, otherwise I die

System Behavior

The system is still not functionally flowing because I am working on modeling the in-flow, care and outflow process.

Rationale for agent rules

Much of this model is based off the APACHE(Acute Physiology and Chronic Health Evaluation II) acuity scoring system used in health care. These scores are heuristics derived from 12 "routine physiological measurements" which are typically taken during the patient admission process. The higher the apache score the more severe the condition. The mortality rate can be inferred from the apache where the predicted mortality increases logistically with apache score.

I will be using the apache score as the primary method for characterizing acuity of patients. The hope is that I will be able to use the apache score as a reference pattern. Since it is well documented I should be able to find distribution information. Acuity in the ICU setting is very important because it can change the workload dynamics. It has an effect on length of stay as well resource utilization. In other words, a rural ICU that does not have many acute patients is very different from Northwestern's ICU where the sickest patients are transported in.

Because acuity plays an important role in the flow process I plan to use it as both a method of reference and calibration as well as a tool to understand probability of morbidity.

Questions

1. How to best model the care process? Should patients have their apache reduced by a caregiver a certain amount each tick? Should patients with high apache be more responsive to care? H

1. Visualization. It seems there is a better method to model the patients. Maybe put them in their own rooms?

Next Steps

1. Implement probability table for patient arrival.
2. Implement mortality probability table for apache scores.
3. Implement care process.
4. Finish caregiver assignment process.
5. Look at discharge