Healthcare organizations perform a multitude of transactional processes, often with sub-optimal performance. Minitab helps many healthcare organizations adopt data-driven problem-solving methodologies to improve transactional processes. I will share an example of how Minitab can help escalate claims transactions using predictive modeling and Monte Carlo Simulation.
Speed Claim Payment Time
The time taken to resolve a healthcare claim and obtain the payment for services rendered is a key measure. Improving the efficiency of healthcare claim transactions can significantly improve revenue and cashflow. Assume that the median amount for a claim is $5200 and the average volume is 1,200 claims processed per month. Interest costs build quickly as each day that a claim is not settled and results in significant costs to the organization, not to mention the strain on cash flow.
The output measure of interest is the elapsed time in days from initiation to payment for health claims. Other factors involved include the workload of claims processed per person per day, daily error rate (missing information, incorrect values, etc..), and cost of processing. Fields in the data note if the claim involved private insurance and if the procedure was considered outpatient.
Related Webinar: Improve Healthcare Services and Patient Outcomes with the Power of Data Analysis
Slow claims cost the organization in many ways including increased interest payments and unreliable cash flows. The strategic business plan included a goal to settle claims within 50 days to maintain healthy financial performance. To meet this goal, we used a capability study to determine how well the population of claims processed meet the goal. As a result, the process can now be expected to produce most claims (883K out of a million) that will process in more than 50 days.
Maintaining Stability of the Claims Process
How can the client ensure they maintain the stability of the claims process over time? There may be special causes that are behind spikes in claim processing time, which could be easy to address. Variation in claim time that is stable over time suggest a more detailed approach to gaining overall improvement. A quality control chart is a great way to assess process stability.
The individual and moving range charts provide evidence that the claims settlement process is stable with an average of roughly 53 days. Claims can vary randomly between 46 and 60 days without being considered out of the ordinary.
Creating a Regression Model
Often, transactional processes involve many variables that might contribute to changes in claim settlement time. Here are some of the best models to use:
- Predictive analytics is an invaluable solution for finding focus with variables that have the most leverage.
- Automated machine learning is easy to use, yet extremely powerful for fitting a useful model.
- The TreeNet model explains 81.8% of the variability in claim time narrowing focus to 3 continuous variables (claims processed, cost of claim, day error).
Once the input variables have been reduced, a regression model was created. The regression model is easy to explain since it is a linear combination of the model inputs. The regression model for the three inputs explains more than 78% of changes in claim time, which is nearly as good as the more complex Tree Net model. Making predictions from the model explain roughly 75% of the variation in claim time as shown below.
Related Webinar: See the Unknown with Monte Carlo Simulation
Using Monte Carlo Simulation to Refine Data
Once the regression model is created in Minitab, a Workspace or Engage user can easily import it into a Monte Carlo simulation. The historical information on the shape and parameters of the 3 inputs variables is entered along with the goal of no more than 50 days of claim processing time. It is easy to use the data to automatically create the behavior of the inputs if parameters are not known.
A Monte Carlo simulation is run for the default of 50,000 iterations obtaining results that indicate the capability of the current process, which is poor as noted above. Utilizing parameter optimization of the mean levels of the inputs is done quickly to get much better results with tuned input levels.
Optimization significantly improved claim time – now less than 1% of claims are beyond the goal. The path to improvement involves reducing the workload to 30 claims per person per day, increasing efforts to reduce errors to no more than a tenth of a percent, and adding resources that will increase the cost per claim to around $345. The increased costs and reduced error will likely come from hiring more processors. A marginal increase in costs is only a small percentage of the benefits enjoyed freeing up revenue through more efficient settlement time.
Parameter optimization is achieved through shifting the average values of the inputs to the levels identified by the optimization algorithm. Additional improvement is possible by analyzing the sensitivity of claim time to variation in the inputs about the mean settings.
Claim time looks to have the most sensitivity to variation in the workload. Stakeholders of the improvement project determined that a 20% reduction in variation can be achieved by having additional manpower in place to better balance the demands over the workweek.
A new run of the simulation reflecting this change provided evidence of gains in capability resulting in a very small percentage of claims (0.04%) that exceed the goal. The end capability value (Cpk = 1.1) indicates that 110% of the distribution will fit within the 50-day maximum processing time goal, which will be robustly met.
Confirm Results with a T-test
Simulations provide evidence on an expected population of results; however, there is no substitute for confirmation data. There are influences that can affect the output that may not be known or difficult to include in statistical models. The sponsors of the claim time improvement project included the collection of 100 random observations as a confirmation sample and used it to determine the capability to settle within the 50-day goal. A two-sample t-test demonstrated the reduction in average claim time is highly significant (p~0.000).
The capability to settle claims within the 50-day goal was greatly improved with less than 300 claims in a million that may exceed goal.
As I hope this blog has demonstrated, modeling and simulation study allow for the study of the process and estimation of the effect of improvements before changes are made. These steps enable stakeholders of the process to gain reasonable assurance regarding the robustness of the suggested changes. For major healthcare organizations, the business value of the improvements is likely to be in the millions of dollars. The Minitab solutions used in this blog were easily deployed by non-statisticians and provide excellent documentation for auditors and for sharing best practices across the organization.