What if you could use data science to determine the best course of action for any given situation?
How much of our lives do we spend contemplating what we ‘should’ do? For example: should I be risk averse and leave for the airport three hours before my flight, or two hours beforehand to avoid wasting time sitting around when I get there? I might start checking the traffic on Google Maps at the three hour mark, or reflecting on what my experience has been with traffic in the past when I leave at such times. Luckily, the branch of analytics called “prescriptive analytics” helps tackle the issue of finding the best course of action in a given situation. One of the most fundamental and powerful tools in the prescriptive analytics space is simulation.
Applying simulation models to reduce hospital operational expenses
Simulation models are becoming very prevalent in the healthcare industry due to the inter-dependency of factors in healthcare operations as well as the wealth of data that exists across the entire industry. Almost any hospital environment can be simulated. This begins with the data scientist developing a thorough understanding of the hospital’s current state process and establishing certain key performance indicators. Once the environment is simulated, the simulated current state metrics are assessed to see if they match the true, observed current state metrics. A should be able to look at the simulated results and agree that they match reality. Once the simulated current state resembles the true current state, the possibilities for measuring the impacts of process changes by simulating a potential future state are endless. To explore this further, we’ll use inpatient pharmacies as an example.
Simulating an inpatient pharmacy environment
A situation with single piece flow where each drug to be administered is dispensed right when the patient needs it would minimize waste and ensure patient safety. In the real world this is difficult to achieve, so drugs are prepared in batches. But how does the pharmacy director decide how many batches are necessary and when they should occur? Suppose a facility has 6 batches scheduled four hours apart. Are those the best times to begin each batch? What if the morning batch windows were shortened and the evening batch windows extended? What if they were each shifted backward by one hour to align with when orders are typically discontinued? Maybe the chief pharmacist wants to expand their facility’s batch schedules from 6 per day to 8 per day. This comes with a significant increase in labor cost, but would it be offset by the savings from waste reduction? Such questions can be taken from the realm of speculation to one of tangible insight using a well-built, validated simulation. Every conceivable schedule of batch windows can be simulated, with drug waste and labor costs accounted for. Thus, an informed decision can be made on a best possible batch schedule without a costly trial and error approach.
Certain metrics associated with waste need to be understood. It could be that once a patient is transferred to a particular ward, or once their encounter has a large number of wasted dispenses, that subsequent dispenses on their medication order are far more likely to be wasted. However, imagine a queue where dispenses that are flagged as “risky” are prepared after dispenses without such risk factors. It is possible that this setup could catch some of these doses before they get wasted and, as a result, save a non-trivial amount of money. Potential “novel” interventions like these can be integrated into the simulation and their effects on waste can be measured, providing leadership with key insights of how to reduce waste costs.
The endeavor of building an effective simulation model for a pharmacy (or any other hospital environment) can be a long one. It begins with fundamental understanding of the facility’s process and collection of robust data that often lives in multiple disparate source systems. Dispense events must be accurately linked to administration events, and the amount of waste and cost attributed to said waste must be well understood. There must then be a marriage of process improvement and data science techniques to ascertain root causes associated with particular waste events. Finally, because measuring improvement requires a baseline comparison, a simulated current state must match the true current state and receive appropriate clinical buy-in before being used to prescribe intervention recommendations. However, this work comes with endless potential and incredibly great reward – applicable not just to inpatient pharmacies looking to reduce costs, but to all hospital departments seeking operational excellence and increased margins.
About the Author
Richard Bryant is a data scientist with experience using data analysis, predictive modeling, and discrete event simulations to drive innovative technology solutions. Utilizing an M.S. in Applied Statistics from the University of Michigan and a Lean Six Sigma Green Belt certification, Richard has provided analytical support to healthcare process improvement projects in various hospital departments, such as radiology, obstetrics, and pharmacy. Prior to his role at Improvement Path Systems, he provided statistical consulting services to the Michigan Surgical Quality Collaborative and to the educational technology provider Echo360.