1. Readmission Reduction

2. Never Event Reduction

---

Readmission Reduction

Readmissions are an integral part of the upcoming change in health care reform. As the financial incentive from readmissions reverse, hospitals need a way to prevent readmission. Many tools in the market today provide historical statistics to ascertain if process change is having the desired effect. Knowing your complication rate for last month or your readmission rate for last month does little for the patients who already had the complication or for those who have been readmitted.
Hospitals need a pro-active tool, not a reactive tool.

Attempts at creating pro-active risk stratification tools for readmissions or for that matter, any desired outcome, are largely based on the best guesses from subject matter experts. In new focal areas such as readmissions, publications are limited and certainly, evidence to support the publications are not based on a long period of time. Thus, the hospital may be adopting methodologies that are largely unproven if not sometimes incorrect in some areas. A common flaw with such methodologies is that in trying to determine the risk parameters, the human mind often mixes up correlation and causation. A parameter that is correlated with the outcome often is bright and flashing, begging for attention to be included in risk stratification models designed without mathematical tools to help differentiate causation. Using an extremely lay-person example, lightning is correlated with rain (when there is rain, chances are that sometimes, there is lightning). But lightning is not the cause of rain.

This simple misstep could lead to a lot of lost opportunity in medicine. Using a medical example, assume a model to ascertain risk scores for VTE is built without mathematical analysis. The human mind may observe that orthopedic surgery units have a lot of VTEs. That does not make the individuals with orthopedic surgery riskier for a VTE. It is a correlated parameter. The underlying causal parameter is mobility. When an individual undergoes certain kinds of orthopedic surgery, they are immobile. Immobility is a huge risk factor for VTEs. Thus, there are many kinds of orthopedic surgery (e.g., wrist surgery) where the patient has little or no risk for VTEs. Likewise, there are many other units in the hospital where a patient may be immobile, making them highly risky for a VTE. Many models in use today for VTEs use correlated parameters, skewing results and using hospital resources in a sub-optimal manner.

Hospitals need a mathematically based tool to manage readmissions, not a heuristics based tool.

Our Value Add

Rulester provides a provides a PRO-ACTIVE, EVIDENCE-BASED, MATHEMATICAL tool to identify patients at risk for readmission before the patient is discharged. We then provide a pathway for the best discharge location (home vs. SNF vs. etc.) and within SNFs, help you scorecard the particular SNF that has the lowest likelihood of readmission for each patient.
We deliver our insights in plain English, usable by all staff and work with you to help you succeed.

Rulester Methodology

Our methodology involves a three-step process: The discovery of early warnings, the institution of remedial actions based on the early warnings and the monitoring of the remediation for effectiveness so the next cycle can be improved.

Foundational to all these steps is the process of setting benchmarks. Let us now understand benchmarking. Put simply, a benchmark is a standard or an expected value against which actual are compared. Rulester will analyze your transactional systems to identify from past history, in some cases incorporating subject matter expert overlays of the past history and finally, use industry comparisons of transactional data where available to construct such benchmarks. To us, a benchmark is not a single high-level number. We may create 20-30 benchmarks for a business. For example, the expected readmission rate for a 50-60 year old female with a primary complaint of chest pain at the ED may be a benchmark of 23%. We may then have several benchmarks by additional diagnosis as they become available, by medications and by body weight, to use the same example.

Setting such benchmarks is generally a very involved process and would involve a rigorous analysis of a single hospital’s data and/or an entire industry’s data if such a repository is available. Hospitals are analyzed anonymously so their identity is protected and industry metrics are compared to a single hospital’s performance to keep the exercise beneficial to all. For example, another benchmark may be that a particular hospital’s readmission rate for 30 and under, male patients admitted with symptoms for pneumonia is trending $2,309,000 ahead of the “market” where the market numbers are based on an industry comparison across hospitals. In this case, we would have converted the impact of the readmission advantage into DRG dollars so performance-based compensation can be managed. In many cases, a hospital may choose to peg their internal benchmarks as a deviation above or below market – for example, a certain hospital wants their readmission of a certain complaint to be 2% below “market”.
Once such benchmarks are set, we then begin the process of unearthing early warnings – which are essentially levers that enable a hospital to respond to deviations from benchmarks before the problem actually is reality. Thus, we enable a hospital to avoid problems before they happen, such problems being defined as missing their internal or market benchmarks.
We will now discuss the early warning methodology in context of three different industries so it may be better understood.

Case Study #1 – Steel Distribution (Chicago Tube & Iron, Romeoville IL)

The challenge – increase the number of quotes that convert into orders while simultaneously maintaining or increasing margins. Our platform is best represented in the illustration below:

The Discovery phase in this example unearths that a certain segment (agricultural customers) are trending lower than benchmark and therefore represent a budget shortfall for a future period. a higher risk of missing comparative “norm” or budget. This early warning may be communicated to specific individuals (e.g., general manager, sales territory manager, etc.). The comparative benchmark may be set based on the data analytics of historical quotes. A predictive model may then be built to construct a score for each quote to convert into an order.

The response to the early warning may be remedial actions that the general manager and outside sales managers may take. For example, conserve resources by not quoting certain customers who are responsible for extremely low orders per 100 quotes, while also incenting others who exhibit a high ratio. In the meanwhile, the internal sales team may optimize their day by working on those quotes with the highest predictive score to convert into an order while getting to the lower scores at the end of the day.

The final leg of the journey is the monitoring phase where our system keeps score of the remedial action. Did the reassignment of resources give a higher yield (more orders per 100 quotes)? Was margin maintained? Did the predictive score turn out to be accurate (i.e., a higher score = higher chance of converting from a quote to an order)?

The result of the monitoring phase is then used to further refine the early warning for the next cycle, improve the predictive model and provide additional information for the general manager and the outside sales person to consult while making their decisions.

This platform is in play at Chicago Tube & Iron in Romeoville IL, a $250M steel distributor. The company aims to make $6M more in profit by improving quote to order ratios by 2%.

Case Study #2 – Hospitality Industry

The challenge – increase the revenue per available room (REVPAR). Our platform is best represented in the illustration below:

The Discovery phase in this example unearths that a certain segment (group bookings) has a higher risk of missing comparative “norm” or budget. This early warning may be communicated to specific individuals (e.g., group booking manager, revenue manager, etc.). The comparative benchmark may be set based on the data analytics of a large set of data of properties in the same market or sometimes even taking data from other markets that are reasonable comparisons (e.g., using large metros with > 3M population as a whole, instead of Chicago only)
The response to the early warning may be remedial actions that the revenue manager may take. For example, the team may decide to market online to attract customers, drop prices, create promotional deals or in some cases, increase prices if elasticity of demand indicates that is the right action.

The final leg of the journey is the monitoring phase where our system keeps score of the remedial action. Did the marketing program work? How many rooms were picked up?

The result of the monitoring phase is then used to further refine the early warning for the next cycle, improve the predictive model and provide additional information for the revenue manager to consult while making their decisions.

This platform is in play at a certain multi-national travel industry player in the Chicago area who has in excess of 7000 hotels as customers. The software and data provider proved that the analytics can add 11% to revenue while adding 5% to profit simultaneously.

Case Study #3 – Hospitals (Advocate Good Samaritan Hospital)

Let us review this in the context of a hospital pharmacy trying to improve the process of when certain medications are administered to certain patients:

The Discovery phase in this example unearths that a certain patient has a higher risk of an adverse event compared to the “norm”. This early warning may be communicated to specific individuals (e.g., the hospital pharmacist, the nurse on the floor, attending, etc.).

The response to the early warning may be remedial actions that the pharmacist and/or the physician take. For example, the nurse may initiate a call to the doctor and suggest a certain drug. The physician may or may not agree to the suggestion and makes the final call.

The final leg of the journey is the monitoring phase where our system keeps score of the remedial action. Did the drug turn out to be effective (prevent complications)? Did the drug have adverse side-effects? If so, what patient demographic was the side effect and where did it not happen?

The result of the monitoring phase is then used to further refine the early warning for the next cycle, improve the predictive model and provide additional information for the physician to consult while making the decision on administering certain drugs.

This platform is in play at Advocate Good Samaritan Hospital in Downers Grove, IL – a top 50 Thomson Reuters hospital and a Lincoln award winner for quality and low complications. The hospital aims to further reduce complications from surgery using the platform.

Readmissions Pilot at Getwell Hospital* (*Name Disguised)

In Summer 2010, we completed a pilot project at Getwell hospital. Whereas the average patient had a 14% readmission rate at Getwell, we identified patients with early warnings of much greater than 60% of being readmitted.

We then analyzed the discharge locations of these patients. Using internal data warehouses and our proprietary analysis tools, we risk stratified patients who were in the hospital as of June 2010 for the possibility of readmission in July 2010. We then went back to the data in August 2010 to see if our model was accurate. The results were stunning. We were able to predict with 4X more accuracy than a random sample, that a certain patient was likely to be readmitted even before they left the hospital.

In addition to that, we then created a SNF scorecard that identified the best SNF a patient could be discharged to, based on their diagnosis/complaint to reduce readmission. We created a methodology to use data from all Advocate hospitals to increase the comparison set. The project is awaiting executive sponsorship to move to the implementation phase. This document is a proposal to move the project further.

Next Steps at Getwell Hospital

More Data – To proceed to the next steps at Getwell Hospital, we would need to add more data to the model. As Rulester starts work with other area hospitals, the opportunity to set the initial benchmark is immense.

Segmentation – The next step to be done is to segment nursing homes. Using more hospital data, we cannot compare a nursing home used by another hospital to one used by Getwell as is. However, we may be able to create a segment such as “50 beds and under nursing homes with ventilator assistance”, and thus classify SNFs across hospitals into one segment. This is a very involved project with tremendous dividends. The resulting classifications will again be fed into the mathematical model to answer questions such as the incremental reduction in readmission by sending a patient to one kind of nursing home vs. another compared to sending the patient home. We will for the first time in the industry, create a comparative score card for discharge dispositions that is scientifically based and Advocate stands to benefit from it on day #1.

Integration

The model will then be integrated into your electronic medical record so the risk stratified score can become part of the data the medical team sees every hour. At a certain hospital, we refresh our model every hour, based on new lab values. We would do the same for readmissions. Thus, a new lab value or a new complaint/diagnosis may change the risk of readmission and we will instantly alert relevant parties of that.
Evidence-Based Feedback – The final step brings everything back full circle. That is to build the evidence feedback loop. Assume the care team worked with the patient and the family to select a certain discharge disposition. Even if the hospital may not be in a position to require or suggest a discharge destination, Rulester may play a third-party role in sharing the analytics with the patient, thus guiding the patient to the best care with the lowest readmission risk. Having done that, the next question is whether the analytics was right. Did that patient get readmitted? If so, and several such occurrences change the analytics mathematically, the model would change real-time when evidence supports such a change. If not, the model would more strongly point to one direction vs. another. Thus, the built-in feedback mechanism would continuously improve the process and the product.

Financial Approvals – The project’s multiple phases and the complexity of the work requires a financial commitment of $350K from Getwell for the first year. The amount would be payable in multiple phases, consistent with milestones to be met.

ROI Analysis

The average cost of a returning patient to a hospital varies widely by the complaint. Often times, the readmitted patient is more complicated than the original admission, driving up costs for the ACO.
Industry metrics on the average cost impact of a readmission vary, but are centered around $3000-$4000 per patient. Since the case mix and the hospital location have immense variability, we will use $3500 as a plug figure for Getwell Hospital.

References:

http://www.ncbi.nlm.nih.gov/pubmed/17672809 – $14K for some disease states

http://circ.ahajournals.org/cgi/content/full/circulationaha;94/6/1350 – Median readmit cost $5842…range $3549-$12,170

http://www.bmj.com/content/333/7563/327.full – $4000

Direct Cost ROI

For this project with Rulester to break even, we would need to drive down readmissions by 350000/3500 = only 100 fewer readmissions a year to break-even! The direct costs of this project are easily offset by benefits.

Indirect Costs ROI

Assume further that the resource commitment from Getwell to intervene in cases (create a SNF follow up program, identify internal focus teams, IT integration, etc.) is $1M. This is another 300 readmissions to be prevented.

Summary

At a hospital with 40000 discharges a year and a current readmission rate of 14%, the project will break even if readmissions are reduced by 1%. Easily achievable with your commitment and our expertise.

Never Event Prevention

VTEs

Rulester’s healthcare platform provides early warnings to reduce never events. For example, our platform provides predictive risk stratification of patients at risk for venous thrombo-embolism and additionally also delivers analytics on prophylaxis, its effectiveness and the risk of complications with and without chemoprophylaxis. The platform is integrated with the hospital’s EMR and therefore keeps current with each admission and discharge. The moment a new patient is stratified as a high risk individual who does not have chemoprophylaxis, we enable the anti-coagulation team to intervene and communicate with the physician. The anti-coagulation team and the physician is also presented with evidence of comparative hospitals for the usage of chemoprophylaxis, the incidence of DVTs without chemoprophylaxis and the resulting complication such as a bleed with chemoprophylaxis. Our system dynamically collects, polls and analyzes data in real-time and thus is able to present point-in-time evidence to the physician and/or anti-coagulation team. The product is being implemented hospital-wide at Advocate Good Samaritan Hospital in Downers Grove, IL.

Slip and Falls

Other hospital products include an early warning system to predict silp and fall risk for patients, presenting evidence for best fall prevention measure by patient type and integrating with the ‘EHR’ so the hospital can compare themselves to others in the area real-time.

Readmissions

We are also in the process of evaluating early warnings for readmissions. Thus, we would score a patient upon admission for readmission risk and enable the hospital, the caregiver and the patient to work as a team with the physician to provide care.