As we try to identify key causes of high US healthcare costs, the topic of variation, specifically variation in practice style, often emerges. It is an accepted fact that in manufacturing quality is defined as the absence of unnecessary variation. In other words, variation deteriorates quality. This leads to mechanisms to measure and assess variation and attempt to eliminate them. In traditional manufacturing, this is a fairly straightforward process, although challenging. In health care this is more challenging since we are dealing with potentially different patients with highly varying health care conditions and multiple approaches to resolve the health care concern. This article will focus on several suggestions to make this process easier.
The highly complex health care system requires at a minimum three basic assessment tools for the reviewer to identify unnecessary or inappropriate variation:
- Accurate health status risk adjuster
- Appropriate benchmarks
- Access to accurate analytics
Accurate Health Status Risk Adjuster
Since every patient is potentially different, it is critical there is a way or methodology to accurately measure and/or assess the patient’s health status and/or their natural demand for health care services. Traditionally we find that this often results in a “risk score” or “risk adjuster”. Today we have a multiplicity of approaches to make these types of assessments. Without some type of risk adjustment, the analysis is flawed. The appropriate selection of a tool to measure and assess risk improves the reliability of the analysis. Medicare Advantage plans utilize hierarchical condition categories (HCCs). The ACA program (“Obamacare”) utilizes an adjusted HCC approach. Vendors have developed proprietary tools (e.g., 3M and their CRGs). All of these approaches review specific patient demographics and diagnostic and cost information and develop some type of risk score or factor to describe the patient’s related risk.
Accepted techniques utilize these risk factors to normalize the results when comparing and assessing patient variation. Perhaps the simplest analysis example involves the analysis of inpatient length of stay. A younger and healthier population would be expected to have a lower average length of stay than an older and sicker population.
Once we have carefully measured and assessed the risk of a particular population, the next task is often the identification of an appropriate benchmark. No benchmark is perfect, however, the establishment of a benchmark is key to this analysis. The historical comment that “the biggest enemy of a good plan is the perfect plan” definitely applies to benchmark determination. It is better to compare to something than never compare to anything. The author has helped determine a variety of benchmarks over his professional career. Most benchmarks need some type of updating over time as they are refined. However, if done right the first time, benchmarks can be valuable over a long period of time.
One of the simplest benchmarks to determine is the average length of stay. The following process has been used over the years to develop meaningful length of stay benchmarks.
- Obtain detailed eligibility and claims information for a particular population
- Choose a type of stay categorization scheme (e.g., APR-DRG, MS-DRG, etc.)
- Sort and summarize actual data by type of stay including the number of stays, number of days, average length of stay
- Exclude special cases from the data
- Transfers in and out (naturally have shorter stays since they spent some of their stay elsewhere)
- Within each type of stay (i.e., DRG) develop a distribution of stays, determine the mean mode, and various percentiles (e.g., 20th 40th, 60th, 80th, etc.)
- Select the length of stay that corresponds to the 20th percentile and calculate an overall average length of stay for the population
This approach helps generate a reasonable best-in-class benchmark for testing purposes. Assuming the DRG categorization scheme in use has reasonably separated out various acuity levels then this can serve as a reasonable benchmark. In other words, people were able to manage that patient at a much lower level. This particular benchmark would be characterized as the “uncomplicated patient” average length of stay or the best you could ever do with the patient with the fewest complications.
Assuming the DRG categorization scheme in use has reasonably separated out various acuity levels then this can serve as a reasonable benchmark.
The above benchmark needs to be adjusted to reflect the reality of complications. For a commercial under-age 65 population studies show that as much as 80% of all patients are uncomplicated suggesting that 20% have complications. For a Medicare population, the 80% uncomplicated ratio drops to about 50%. A Medicaid population has 60% – 70% uncomplicated.
Going back to the distribution of stays by DRG, you then calculate the average number of days for those patients above the 20% threshold used earlier. The total benchmark becomes the sum of the uncomplicated patient’s average length of stay plus X% of the excess days above the uncomplicated patient’s average length of stay where X% is the percentage complicated as discussed earlier.
We have applied this approach and compared it to best practices from our clients and have found that this works well to develop preliminary benchmarks for high performance.
Access to Accurate Analytics
The third step in the process is to compare and analyze actual performance. Staying in the average length of stay analysis would involve the analysis of the actual length of stay to the benchmarks previously developed. In the case of inpatient stays, we need to consider both the length of stay differences and the distribution of stays by DRG. As systems become more efficient more and more patients are treated in an outpatient setting which can change the underlying distribution of stays. Both need to be considered. As more and more analyses are completed characteristics of each system can be compared to results to determine what factors and care management practices are associated with the most efficient stays.
As variation in care is minimized, the quality of care increases. We are not suggesting a “cookie cutter” type of medicine but rather taking advantage of processes that work in one setting to more areas. Ironically significant variation occurs within a single facility or medical group. There are regional variations, but more importantly, there are substantial variations by provider.