Insight
Simulation for Predictive Analytics in Healthcare: Answering “What If?”

>>Simulation for Predictive Analytics in Healthcare: Answering “What If?”

Using simulation as part of healthcare data analytics is a powerful, low-risk, and low-cost approach to figuring out the best method, system, or decision for your clinical and business objectives.

As part of your data analytics journey, we believe that simulation offers a vigorous way to test out variables and potential solutions or changes to a system without increasing patient risk, wasting precious dollars on untested pilots, and expending too many necessary resources on projects that aren’t vetted.

This article provides an overview of simulation, the benefits, and a real-world use case for simulation in the healthcare domain to reduce wait times and optimize resources.

What is Simulation for Predictive Analytics?

Simulation is the operational imitation of a real-world process or system over time. In your simulation journey, it fits partially between the predictive and prescriptive phases.

Traditional Data Analytics Journey Chart

Start by using human knowledge of cause and effect to create a model of the system where the problem exists. Then, connect the available data to that model and simulate scenarios and future projections based on data, variables, and situational parameters.

Simulation is useful when you want to:

  • Study and experiment with complex systems and internal interactions without the cost of implementation
  • Simulate changes to observe effects
  • Suggest improvements to a system
  • Derive variable importance and demonstrate how variables interact
  • Experiment with new designs
  • Visualize a system
  • Introduce soft factors like time pressure, morale, and reputation

The Benefits of Simulation in Predictive Analytics

In healthcare, anticipating needs, driving operational efficiencies, and choosing the optimized option is not a simple question of supply and demand. Interactions with patients and members are complex and variable.  The quest to improve is consequently difficult, sometimes costly, and potentially risky – experience and outcomes may suffer.

Simulation is a way to test hypotheses without risking patient safety and incurring costlier methods of analysis.

It can be used to accurately and comprehensively model anything from a single clinic to a network of hospitals. Financial controls, human performance factors, patient activities, staffing variances, as well as constraints that have yet to be realized can be added, edited, and expanded with ease.

Lastly, simulation lets you test various business rules that will affect your case; developing a prescriptive model that is bounded by quantified results.

Types of Simulation in Predictive Analytics

There are several types of simulation used frequently in healthcare. Discrete Event and Agent-Based are the two, main types. Monte Carlo simulation can be incorporated within both of these types as a means to incorporate randomness into computations.

Discrete Event Simulation (DES)

A discrete event simulation models systems as networks of queues and activities where state-changes in the system occur at distinct points of time. Traditionally, it’s been used at an operational or tactical level to answer specific questions. In healthcare, those domains include solving resource allocation problems and patient wait times.

According to the Journal of the Operational Research Society, a discrete event simulation (DES) “is a method of simulating the behavior and performance of a real-life process, facility, or system. DES is increasingly used in healthcare services because the technique can be applied to problems of greater complexity and scope.

DES models the system as a series of ‘events’ [e.g., a birth, a stay in an intensive care unit (ICU), a transfer or a discharge] that occur over time. DES assumes no change in the system between events.” [citation – Katsaliaki K, Mustafee N. Applications of simulation within the healthcare context. J Operational Res Soc 2011;62:1431–51. 10.1057/jors.2010.20.]

Agent-Based Model (ABM)

An agent-based model is a type of computational model for simulating the actions and interactions of autonomous agents – individual or collective organizations or groups – and analyzing their effects on the entire system. ABM combines elements of game theory, complex systems, emergence, computational sociology, multi-agent systems, and evolutionary programming.

For both of these types of simulation, Monte Carlo methods can introduce randomness.

Monte Carlo Simulation

A Monte Carlo simulation is a computational algorithm that relies on repeated random sampling to obtain numerical results, using randomness to solve problems. Typically, it models the probability of different outcomes that aren’t deterministic. They can’t be predicted due to the intervention of, or dependency on, uncertain variables in a process. In healthcare, it might be applied to improve operations, such as ED-to-inpatient bed demand or OR-to-PACU, for example.

Because Monte Carlo is a computerized mathematical technique, it can account for risk in quantitative analysis, offering a wide range of outcomes and the probabilities that will occur in light of changing variables. It also demonstrates the likelihood of extreme possibilities — the most conservative decision and a more middle-of-the-road choice.

Figure 2, below, shows various distributed outcomes for one’s retirement income and the most likely outcome among those simulated futures.

According to an article in Becker’s Hospital Review, “Using multivariate statistical analysis and appropriate modeling, healthcare organizations can go beyond binary analysis to show all the possible outcomes of clinical treatment, insurance premium pricing, business valuations, and even personalized medicine in near-real time.”

Real-World Use of Simulation for Predictive Analytics in Healthcare

Problem:  

The wait times experienced by patients in hospital outpatient departments is a problem seen everywhere. Outpatient departments face increasing pressure to improve the quality of their services and patient experiences by reducing wait times.

Figure 2 - Monte Carlo Simulation Example
additional staff

An outpatient clinic under study sees an average of 100 patients a day.

open 9 to 5

The clinic sees patients from 9 am to 5 pm and the arrival time between patients is five minutes.

The clinic employs:

nurses

Two nurses (who have 15 minutes per patient).

more doctors

Three physicians (who have 20 minutes per patient).

Patients are complaining about long wait times before being seen by staff, and they are sharing low ratings online because of having to wait so long.

Approach:

Two possible changes could reduce the wait times:

doctor patient

Add staff so that they can see more patients simultaneously or more quickly.

Improve efficiency of existing staff via workflows, office layouts, better technology, or training.

Because the effort and logistics to experiment with these potential changes would be expensive, disruptive, and may cause further patient dissatisfaction, we built a discrete event simulation model to study the effects of both types of potential strategies to reduce wait times.

Simulation & Solution:

The initial step of the simulation was to model the current state.  This is shown in Figure 3, indicating a queue of 18 patients and wait times of up to 160 minutes.

The simulation for adding staff versus improving efficiency (details below) showed a very acceptable queue and wait time, resulting from adding one nurse and one doctor.

In contrast, moderate cycle-time reductions for nurses did not make a significant impact on the queue or wait time.  Changing the doctor’s cycle time was not explored as many patients view their time with the physician as already limited.

At this point in the simulation, a decision could be made to add staff – if financially feasible.

If adding staff is not feasible, or if the risk of patients “voting with their feet” is tolerable, we would suggest further exploration of cycle-time simulation. The goal would be to determine the “minimum time per patient” needed by the nurse and the administrator to see all patients promptly. Also, we could address the expected variability.  For instance, most people may need only five minutes with the nurse, whereas a portion may need 10.

A Monte Carlo simulation could be incorporated to explore this type of variability and understand the effect on the resultant wait time.  Once better understood, this could also be factored into scheduling so that ample time is allowed for the system to catch up with itself during the day.

Now, let’s see these potential strategies in action.

Hypothesis #1: Add Staff

Two staffing changes were simulated:

add nurse
  • Add one nurse
    • This strategy reduced the queue to 13 patients and cut the wait time to 130 minutes. (Figure 3)
add doctor
  • Add one more physician
    • This strategy resulted in 15-minute wait time and almost no queue. (Figure 3)
Figure 3 Reduce Outpatient Clinic Wait Time Simulation

Hypothesis #2: Improve Efficiency

Two efficiency changes were simulated using the original staffing plan, focusing on using a combination of improvements to reduce cycle time.  In practice, this could be an “a priori” exercise to define a target for improvement or it could be an evaluation of a list of specific improvement activities.

Two efficiency changes were simulated:

15 to 10 reduction
  • Reduce the nurse cycle time from 15 to 10 minutes.
    • This strategy reduced the queue to 12 patients and the wait time to 125 minutes.
  • Reduce the administrator cycle time to two minutes and maintain the nurse and doctor cycle-times at the original level.
    • This strategy held the queue at 16 patients and the wait time at 150 minutes.

In conclusion, simulation enables choices and their impact on outcomes, experience, access, and other key healthcare measures. Exploring strategies and results “on paper” alleviates the time and cost of making actual changes.  While it may not eliminate the need for some “real life” experimentation, it can dramatically reduce cost and resources, as well as minimize operational disruption or patient inconvenience and risk.

With a modicum of data and access to key stakeholders who can identify and characterize realistic change options, you can simulate relatively complex scenarios effectively.

Once a simulation is built, it can become a “digital twin” much like a flight simulator keeps passengers safe. A clinic, unit, or system can be constructed digitally, and simulations re-run as new information and variables come into play.

Moreover, as healthcare organizations experience the need to change more quickly, simulation can be a crucial tactic to help an organization evolve from predictive to prescriptive in their data analytics journey.

null

Insight Author

Naveed Ismail, Senior Consultant

Jeannine Siviy

Insight Contributor

Jeannine Siviy, Director of Healthcare Solutions

Insight Contributor

Priyanka Kadam, BI Consultant