Hospitals and health systems have entered a new era in which their survival will hinge on the extent that they are able to analyze data to identify and realize opportunities for improved performance.
As the dynamics of the healthcare environment rapidly shift to a focus on value, hospital executives are compelled to focus constant attention on improving the coordination and quality of patient care while reducing its cost. Whether these efforts are focused on fixed or variable costs or direct and indirect costs—or all such costs—reducing healthcare costs is a perennial challenge, often exacerbated by fragmented data resources. As organizations respond to the imperative to use technology and data analytics to support qualified decisions and improve care, they encounter myriad questions regarding how to apply systems efficiently and effectively and redesign care management models to optimize value.
Many organizations are shifting attention away from traditional hospital care management models to patient-centric, integrated care coordination that spans the care continuum, including primary, acute, and post-acute care. The new models are focused on decreasing clinical variation, improving care coordination and population health, and optimizing the value quotient (better quality/better cost). Organizations require data analytics to support these efforts.
Simply put, to succeed in the evolving environment of value-based care, healthcare organizations must move forward deliberately with efforts to develop sophisticated data analytics capabilities that make use of the data they have to identify and analyze improvement opportunities and guide the steps required to realize those opportunities.
Key areas on which a health system’s performance improvement team should focus when applying these data analytics include accuracy of medical necessity reviews, effectiveness of high-risk admission screening, timely discharge planning, active care coordination, tight communication in handoffs between episodes of care, efficiency in throughput and patient flow and high value patient care delivery.
Eliminating excess patient days (waste) and reducing clinical variation, in particular, have emerged as two primary objectives for any value-focused initiative. Here, therefore, we present a range of data analytical techniques focused on length of stay (LOS) and clinical variation that health systems can implement today to reduce costs and improve care.
Fortunately, the data required for an analysis of excess day reduction is easily obtainable for most organizations. A health system’s electronic health record typically will contain information for any inpatient on the number of inpatient days and DRGs. And for any given DRG and year, the corresponding case mix index (CMI) and expected LOS calculated by the Centers for Medicare & Medicaid Services (CMS) can be obtained.
To illustrate, the exhibit below presents the average LOS (ALOS), CMS’s geometric mean LOS (GMLOS), the ratio of the two, and the MS-DRG CMI for one hospital’s fictitious sample of patients over a one-year period. a It is evident in the graph showing the trend in ALOS and GMLOS that ALOS has been above the GMLOS for this sample for the past 12 months, indicating a potential opportunity to reduce the LOS for MS-DRG.
Sample Analysis: Average LOS Compared With Geometric Mean LOS
Benefit opportunity assessment. The key is to accurately identify the benefit opportunity. For instance, based on the previous example, given the potential excess days in the sample shown in the exhibit below, and assuming a cost per day of $400 for each day over the expected LOS, potentially $16.9 million in excess expense could have been avoided in this one-year period if every patient had been discharged exactly according to CMS’s expected LOS. Because CMS payments are tied to the expected LOS, excess or avoidable days add cost to the hospitals without corresponding revenue.
Sample Report: Analysis of Potential Savings From Reducing Average Length of Stay (LOS) in Excess of CMS’s Geometric Mean LOS (GMLOS), CY 16
In reality, some patients will have conditions that are more severe than indicated by the DRG. Perhaps the DRG has been miscoded, and it may not be possible to address every single patient and reduce every individual LOS. Executives may decide that a reasonable goal could be addressing 30 percent of the excess days versus 100 percent, over a specified period. The exact mix of patients won’t return to the hospital in the following 12 months, and unless there is a significant change in mix over time, the hospital can track the ratio of ALOS to GMLOS over time to track any improvement, while also comparing the remaining potential excess days or dollars with those in a baseline time period.
Best practices for improving LOS include daily interdisciplinary care rounds to reduce unwarranted excess days from the system, transitioning patients from the acute care stay, and leveraging skilled nursing facilities, home health, rehab, long-term acute care hospitals, outpatient services, and other levels of service.
Analysis of excess cost. A potentially beneficial step when studying LOS is to review groupings of direct costs and excess costs associated with LOS greater than CMS’s GMLOS. For example, in the exhibit below, bubbles are placed in the chart for each DRG indicated, where the X-axis represents the average direct cost at the patient level and the Y-axis represents the number of excess days (inpatient days above the expected GMLOS). The size of the bubble reflects the number of patients in each DRG.
This representation of data is visually appealing because the outliers having high direct costs and high excess days are immediately apparent in the top right quadrant of the graph. Moreover, by choosing a bubble for a particular DRG in this chart and displaying other dimensions of performance, such as ordering physician instead of DRG, executives can review outliers by physician within each DRG, thereby using hospital data not only to reduce LOS but also to address unnecessary clinical variation.
Sample Analysis of Average Direct Cost for MS-DRGs Compared With Average Length of Stay (LOS) in Excess of CMS Expected LOS (i.e., Geometric Mean LOS)
Scorecards and performance KPI analyses. To obtain a broader prespective, executives can use scorecards to identify other high-value and high-target performance improvement areas, as shown in the exhibit below.
Sample Scorecard: Patient Flow Metrics and Readmissions
Another tool for measuring the level of improvements achieved is a report that shows the LOS key performance indicators (KPIs) for a baseline period as compared with the current period. Such a report can clearly depict whether the hospital has improved and to what extent over the designated period. It also is important to note that the exact mix of patients will not be repeated, as volume will likely change, making it preferrable to review volume-adjusted metrics. The potential excess days per discharge (i.e., volume adjusted) for the baseline is 1.7841. For the more current month in the actual time period, the same metric has been reduced or improved to 1.6237.
Comparison of Data Under Performance-Improvement Initiative With 1-Year Base Line Data
Focus on Clinical Variation
Efforts to reduce clinical variation involve assessment of different services and procedures and their costs, and how much variance in cost there is for the same condition or diagnosis within the same system. The differences derive from various factors, including the overuse or underuse of a healthcare service by certain healthcare providers and specialists. Other influencing factors may include patient demographics, comorbidities, severity of illness, risk of mortality, and intensity of service. By focusing on increasing the consistency of patient care delivery within a health system, variation can be decreased, resulting in improved quality and reduced costs.
Although hospitals typically have various levels of cost data available for a clinical variation analyses, data availability often poses a limitation for organizations attempting to analyze clinical variation. The basis for the analysis can range from actual patient level data all the way down to actual charge items, depending on what data are available. Some hospitals can analyze actual charges and costs for items, whereas others can analyze only charges for items and are limited in their costing ability to analyzing costs according to department or revenue cost average (i.e., using a step-down methodology that allocates cost based on a ratio of cost to charges [RCC]). Any method that can provide only an allocated estimate of direct cost is not suitable for clinical variation analysis. In general, the deeper the dive into cost data provided, the better the analysis.
Analysis of direct costs. A hospital’s direct costs are those costs directly associated with patient care. Typically, organizations can have a huge variation in such costs, even for similar patients with a similar diagnosis and similar procedure. Hospitals all too often are unaware of how much variation exists across patient populations with the same illness, procedure, or all-patient-refined (APR) DRG. b
It is in the variations in utilization and delivery that potential opportunities for reducing cost and improving population health exist. One way a hospital can identify such opportunities is by analyzing different levels of direct cost using data from various sources combined into a single report—for example, by comparing direct costs at the patient level. However, the usefulness of data depends on departmental, service line, and organizational effectiveness. Any data-driven project will benefit from a structure designed to ensure that goals are clearly understood, communicated, met, and sustained. Although providing guidance for creating such a structure is beyond the scope of this article, the need for such a structure should be well understood.
Segmentation analysis. Health system executives can use a segmentation analysis to determine where—i.e., in which segments of care—efforts to reduce direct cost are likely to provide the most value over time. Such an analysis, depicted in the exhibit below, enables executives to view variations in direct costs across a variety of dimensions. The exhibit shows, for a sample hospital, the number of discharges, the total direct cost, and the standard deviation in direct costs for several APR DRGs over a six-month period, where there were at least 50 discharges per procedure within that timeframe. The sample includes only procedures with severity codes 1 and 2 as a risk-stratification approach, because including the most severe cases cause the standard deviation in direct costs to be much higher. Ideally, the segmentation analysis also would allow executives to look at each severity-adjusted APR DRG by attending physician, activity code, and various other fields within the database.
Comparison of Average Direct Cost With the Standard Deviation in Costs for 13 APR DRGs, 6-Month Period
The exhibit shows a wide range of both average direct costs and standard deviations within the DRGs shown. Note that smaller-volume APR DRGs in the exhibit show a tendency toward having a higher variance (i.e., a wider standard deviation from the average direct cost per procedure). Such information could provide a starting place for identifying potential opportunities to reduce variation in direct costs at the APR DRG level. In the middle of the chart, for example, the two DRGs for dorsal lumbar fusion both have high means and high standard deviations, as well as some significant volume, suggesting this area might be a good place to drill down into the data and investigate the cause of the variation.
More detailed analyses. On a more detailed level, analyses of charge-level direct costs for specific procedures, accounting for various factors such as number of patients and number of items used, can disclose opportunities to reduce clinical variation.
For example, the exhibit that follows displays direct costs by number of patients for sample blood cultures taken within the DRG of Septicemia over a one-year period. The X-axis represents the dollars of direct cost and the Y-axis is the number of observations or patients. As shown, more than 400 of the roughly 500 procedures cost $37.30, which represents the mode in this sample (the mean is a slightly lower $36), while very few procedures show a cost exceeding $37.20, indicating that there is no significant clinical variation associated with this particular clinical activity.
Sample Analysis: Numbers of Patients by Direct Cost, Blood Cultures Within DRG 870 (Septicemia or Severe Sepsis With MV 96+ Hours)
In contrast, the exhibit below shows a sample of costs for charge items associated with a procedure. For most individual procedures (shown by the individual bars), only one of the items was used, but in several instances six items were used, and in one, the items numbered seven. More noteworthy is the range of costs associated with the usage.
Sample Analysis: Numbers of Items for a Procedure by Direct Cost
This range of costs and the various numbers of items point to a high degree of clinical variation: There is no clear pattern or consistency within the data, and there is significant dispersion about the mean, with relatively high standard deviation. The data suggest a deep investigation should be undertaken to analyze the drivers of the clinical variation.
Analyses such as those described here are likely to uncover many patterns of clinical variation, and some judgment is necessary to determine which subsets of data should be measured for this purpose.
Other metrics and KPIs can be examined to select target areas. For example, significant opportunities for improvement can be identified using a simple chart showing the top 10 direct costs by activity or revenue code. Another effective way to analyze practice variation among physicians might be to review the highest-volume activities or revenue codes by physician.
Bubble charts are particularly useful for detecting anomalies and outliers related to clinical variation. The sample bubble chart in the exhibit below shows, by physician, the average direct cost for a particular procedure on the X-axis compared average units per patient on the Y-axis. Each physician’s overall cost for the procedure is indicated by size of the bubbles. Using this type of exhibit, a few clear outlier physicians can be spotted that have average direct costs and in some cases higher units than average. Treatment of outliers can include investigation and even possible removal from the analysis.
Hypothetical Comparison of Items Ordered and Average Direct Cost and by Physician
Impact of Reduced Variation
As hospital executives analyze opportunities to significantly reduce clinical variation associated with large variances in direct costs, they also should consider the effects of reducing variation in direct costs on the organization’s contribution margin. Therefore, once they are capable of defining and measuring clinical variation, they also require a standard means for analyzing that variation and determining the potential impact from reducing variation on the hospital’s direct costs and contribution margin.
The exhibit below illustrates the preliminary step in such an analysis. This hypothetical example shows risk-adjusted charge-level direct costs for items used, and the numbers of items used at each level of direct cost, for patients admitted to the hospital under a specific DRG in the most recent 12-month period. The objective of the larger analysis is to identify the potential reduction in direct costs that a hospital can reasonably be expected to achieve by reducing clinical variation for a similar mix of patients admitted to the hospital in a subsequent 12-month period. Stated otherwise, the results aim to show how much savings a hospital might have realized if it could have delivered care to the same set of patients represented by the data but with reduced clinical variation.
Sample Analysis 2: Numbers of Items for a Procedure by Direct Cost
The next step is to apply a methodology to analyze the potential impact from reducing the direct costs. Again, such a methodology could be used for analysis of patient-level direct costs or charge-level direct costs, and some judgment would be needed based on the homogeneity of the data and the purpose of the analysis.
There are different approaches to analyzing potential reductions in direct costs. Some executives examine a uniform percentage reduction in all direct costs no matter what size or magnitude, and others prefer an approach in which only the larger costs (i.e., those above the mean) are reduced, which assumes that there is not as much opportunity or ability for the hospital to reduce costs that are already small. Another option is to implement a blended approach that allows for percentage reductions in costs that are different above and below the mean. For purposes of this analysis, the percentage reduction in costs above the mean refers to reducing only the excess portion above the mean—i.e., bringing the cost of items that have higher costs than the mean down to the mean cost.
In determining what percentage reduction in direct costs would be most viable for a hospital, it can be useful to test various percentages to get a sense of the impact on results. Smaller reductions in direct costs also are possible, but they will produce a smaller magnitude impact. The methodology aims to approximate what a hospital can actually accomplish, and the appropriate method should be a reasonable proxy for the type of reduction that a hospital could achieve over the one-year period. The results need not be exact; rather, they should simply give an idea of what a hospital could be expected to achieve overall from reducing direct costs over a specified timeframe.
The methods described here are straightforward and have the advantage of requiring few assumptions. It also should be noted that other reasonable methods exist. Another approach might be to choose a cohort of selected physicians and examine the impact of reducing other direct costs to the target level of that cohort. That said, a full discussion of all methods is beyond the scope of this article.
The exhibit below shows the results of three methods relative to what is shown in the previous exhibit. Method A looks at a reduction of only those costs above the mean, Method B represents the blended method, and Method C looks at a reduction of all costs by a uniform percentage. For the purposes of this example, the exhibits show a 25 percent reduction in excess costs above the mean in Method A, a 20 percent for costs below the mean in Method B, and 20 percent for the reduction in all costs in Method C.
Comparison of Methods for Determining the Focus and Assessing the Impact of Efforts to Reduce Direct Costs
Not All Clinical Variation Can Be Eliminated
Comparing the results of the three methods, Method A produces the lowest potential savings because it is focused only on reducing the excess costs above the mean by 25 percent. Meanwhile, Method C, which reduces all costs by 20 percent, gives the highest potential savings.
It’s important to select a realistic factor and not overestimate the results that can be achieved at the hospital. Depending on the hospital’s resources, several identified target areas could be managed in an upcoming year by dedicated case managers and staff education. The number and potential magnitude, of course, depends on factors such as available resources and the size of the potential impact relative to the effort needed to reduce the costs.
Because of the unique characteristics of each patient, not all clinical variation can be eliminated. However, evidence-based and data-based approaches to reducing clinical variation can enable providers to realize significant reductions in unnecessary or unwanted variation.
Data Analytics: A CNO Perspective
The importance and value to health organizations of implementing comprehensive data analytics to support clinical team efforts in reducing clinical variation, improving care coordination, and raising the quality of care delivered to patients is an important undertaking that can deliver tremendous value to healthcare organizations. One organization that has implemented such data analytics is Centura Health, Denver. Commenting on her organization’s experience, Linda Goodwin, group chief nursing officer, has the following advice for other organizations that are contemplating embarking on such an effort:
“One recommendation for hospitals and health systems adopting a data analytics platform would be to clearly identify the key stakeholders and data consumers that will best drive the actionable data and alignment with effective workplans.”
After reviewing the potential impacts from clinical variation and LOS initiatives, the process of implementing changes begins. To be successful, implementation efforts should begin with the development of specific action plans for the top areas targeted for reduced LOS and clinical variation. Specific, attainable goals should be developed and shared with the key stakeholders across the organization. Regular follow up and monitoring is required to ensure that goals remain on track and readjusted if necessary.
A hospital’s processes for improving the quality and reducing the cost of patient care should be continuous and ongoing, and addressing unwanted clinical variation and reducing LOS are among the primary means for achieving those improvements.
In any care management transformation initiative focused on reducing clinical variation, improving patient care, and lowering cost reduction, data transparency is imperative. The goal is not simply to cut costs; ultimately, it is to create a highly reliable health system in which efforts to deliver value are proactive and collaborative, and in which data are effectively used as a strategic asset.
In summary, hospitals and health systems can best meet their strategic goals for value-focused care, including reducing excess LOS and clinical variation, by adhering to the following structure used by leading organizations:
Step 1: Define. Key metrics and targets of success are identified and defined.
Step 2: Measure. Clinical variation and overall direct costs are measured for selected target segments after a review of the areas with the largest potential impact on cost reduction.
Step 3: Analyze. The impact of a potential reduction in variation and direct costs is estimated.
Step 4: Implement. Action plans are created and interdisciplinary teams established throughout the hospital to achieve reduced variation.
Step 5: Control. Direct costs are monitored on an ongoing basis throughout the project, and variation is compared with the baseline to ensure that both the variation and direct costs—particularly in selected target segments—are both being effectively reduced.
Laurie Jaccard, RN, founder and president, Clinical Intelligence, LLC
Sharon Carroll, FCAS, director of analytics and actuary, Clinical Intelligence, LLC
a. A DRG’s GMLOS refers to the LOS that CMS has determined should be expected for that particular DRG.
b. APR DRGs expand on basic DRGs by including four subclasses of patients that reflect differences among patients in terms of severity of illness and risk of mortality.