Commitment Beyond Numbers Monograph October 2013 ABOUT THE AUTHORS Lee W. Knepler, ACAS, MAAA, Robert J. Walling, III, FCAS, MAAA, CERA pinnacleactuaries.com Lee W. Knepler ACAS, MAAA Lee Knepler is a Consulting Actuary with Pinnacle Actuarial Resources, Inc. in the Bloomington, Illinois office. He holds a Bachelor of Science degree in Actuarial Science from the University of Illinois. He has worked in the property/casualty insurance consulting industry since 2006. Mr. Knepler is an Associate of the Casualty Actuarial Society (ACAS) and a Member of the American Academy of Actuaries (MAAA). Robert J. Walling, III FCAS, MAAA, CERA Rob Walling is a Principal and Consulting Actuary in Pinnacle s Bloomington, Illinois office. He is a Fellow of the Casualty Actuarial Society (FCAS), a Member of the American Academy of Actuaries (MAAA) and a Chartered Enterprise Risk Analyst (CERA). Rob has been a part of the insurance industry since 1989 and a consultant since 1997. He holds a B.S. in Secondary Mathematics Education from Miami University. For more information on this topic, contact Lee at lknepler@pinnacleactuaries.com or Rob at rwalling@pinnacleactuaries.com A Convergence Health care risk analytics is at the confluence of two megatrends of the last decade health care risk and risk analytics. Analytics can be used to not only predict health care risks, but to reduce or avoid them. Risk analytics - statistical tools that use Big Data to reduce, mitigate or predict risk - have impacted almost every aspect of our lives. Real time tracking of spending behaviors, telematics in automobiles, and apps and devices that provide constant tracking of an individual s medical condition are examples that are driving the explosion of available data and resulting analytics that are used in a variety of settings. Similarly, health care risk has been a significant national priority. A widespread effort to reduce the number of negative patient outcomes has led to a wide variety of loss prevention and educational initiatives. One major initiative has been incentives through Medicare for health care providers that utilize electronic health records (EHRs) to improve communication between providers for a common patient. The Affordable Care Act (ACA) is the latest effort to improve the quality of health care, reduce negative patient outcomes, increase access to health care, and reduce the overall costs of health care. Another major change in health care is the amount and types of health care data that are becoming available to the general public. The largest and most extensive data source is the Hospital Medicare Licensee database, also known as the Hospital Compare database, compiled by the Centers for Medicare & Medicaid Services (CMS). This free database collects valuable information about all Medicare-certified hospitals across the country. The intent of the database is to provide patients a better source in making health care decisions. These quality care scores also form the basis for Medicare s outcome-based compensation initiatives. These compensation incentives will influence the way hospitals operate and how they approach their preventable risks. The value of this database is not limited to patient evaluations and Medicare compensation plans. Our research has shown that the information contained within the Hospital Compare database can be a powerful tool in identifying characteristics that materially differentiate hospital professional liability (HPL) costs between hospitals. This ability to differentiate expected HPL costs is a valuable tool to a variety of HPL insurance stakeholders including insurance and reinsurance companies, hospitals, and hospital-owned captive insurance companies and risk retention groups (RRGs). In this research report we have analyzed data from the Hospital Compare database to examine the impact certain quality of care characteristics have on hospital HPL costs. By extension, factors that reduce/ increase HPL insurance costs also reduce/ increase the frequency and severity of negative patient outcomes. The Hospital Compare Database The Hospital Compare database contains information regarding the quality of care for Medicare-certified hospitals. Found on Medicare s website, www.medicare.gov, the Hospital Compare database produces quality scores that can be used to compare the quality of care at hospitals in order to provide patients a better source in helping them make decisions regarding their health care. The quality care scores also enable hospitals to perform peer hospital benchmarking. According to Medicare s website, Hospital Compare was created through the efforts of the Centers for Medicare & Medicaid Services (CMS), in collaboration with organizations representing consumers, hospitals, doctors, employers, accrediting organizations, and other Federal agencies.
The Hospital Compare database contains information pertaining to hospitals including: Timely and effective care How often and how quickly each hospital gives recommended treatments for surgical patients and for certain conditions like heart attack, heart failure, pneumonia and children s asthma. Readmissions, complications and deaths How each hospital s rates of readmission and 30-day mortality (death) rates for certain conditions compare with the national rate. How likely it is that patients will suffer from complications while in the hospital. How often patients in the hospital contract certain serious conditions that might have been prevented if the hospital followed procedures based on best practices and scientific evidence. Use of medical imaging How a hospital uses outpatient medical imaging tests (e.g. computed tomography (CT) scans and magnetic resonance imaging (MRI)). Survey of patients experiences How recently-discharged patients responded to a national survey about their hospital experience. For example, how well a hospital s doctors and nurses communicate with patients and how well they manage their patients pain. Number of Medicare patients How many people with Medicare have had certain procedures or have been treated for certain conditions at each hospital. Medicare payments: Information about how much Medicare pays hospitals It is also noteworthy that the database contains information on the occurrence of several of the so-called never events. A list of never events or Serious Reportable Events (SREs) has been developed by the National Quality Forum and can be viewed as the kind of mistake that should never happen, such as giving medication to the wrong patient, operating on the wrong side of a patient s body, operating on the wrong patient, or leaving a foreign body in a patient. There are currently 29 SREs. The database is free and hospitals are required by the government to report data to CMS. All of the data is updated on a quarterly or annual basis, so the information is timely. In addition, the website maintains historical data going back several years. We reviewed the data within each of these categories available as of September 2012. Since that date, additional information has become available through the website. This data was not included due to the timing of our research. Most of the additional data is related to linking quality measures to the issuance of Medicare payments to hospitals. The additional data shows that the Hospital Compare database is continuing to grow and expand, and should increase in usefulness over time. The Approach We developed a database that combined several variables from the Hospital Compare database with the ground-up loss experience of a major HPL reinsurer. While the Hospital Compare database contains an extensive amount of information, it lacks historical HPL claims and exposure data. The reinsurance company data allowed us to match the experience of approximately 600 unique hospitals over ten report years to the Hospital Compare database. Differences in hospital naming conventions between the two databases presented a challenge. The data for some of the characteristics was in a very granular format that required us to form groupings of individual data levels. The categories for each variable were determined based on mathematical optimization techniques. The historical data was also stratified into different loss layers to better meet the reinsurer s needs. While we found that there were different risk identifiers in the different loss layers, there were meaningful risk differentiators even in the highest excess layers. We then performed a series of multivariate analyses using Generalized Linear Modeling (GLM) on the combined database to form the basis for a pricing algorithm and scorecard for the reinsurer. These analyses explained significant differences in the HPL loss experience of hospitals in the combined database. Differences in both claim frequency and severity between hospitals were identified. Many of these characteristics have been recognized by HPL underwriters and risk managers as impacting HPL claims experience; but the significance of the Hospital Compare database as a health care risk analytics tool is that it can quantify the impact of these characteristics. This can provide underwriters and risk managers with a tool, such as a scorecard, that allows them to isolate the risk profile of individual hospitals or individual risk characteristics that are having a material impact on their HPL claims experience. By utilizing the Hospital Compare database and predictive modeling techniques, we were able to identify a number of characteristics that explain a significant amount of the difference between risks. The analysis also proved to be useful in identifying risk differences related to large losses that would impact HPL excess insurers and reinsurers. In the end, we found 20 variables from the database that when analyzed in conjunction with a reinsurance HPL claims database showed predictive power. These 20 variables were significant to varying degrees, but we found that each one added value in determining the relationship between the Hospital Compare quality scores and the HPL claims. The results of our analysis for several of the risk categories are contained in the following sections. Timely and Effective Care Many variables related to timely and effective care were found to be valuable in the models we created. These variables are directly related to the quality of care that patients received during their stay at a hospital. In general we found that the worse the quality score, the higher the indicated frequencies and severities of HPL claims. The following is a partial list of the variables available through the Hospital Compare database with regards to timely and effective care: PINNACLE MONOGRAPH 2
Average number of minutes before outpatients with chest pain or possible heart attack who needed specialized care were transferred to another hospital Outpatients with chest pain or possible heart attack who got aspirin within 24 hours of arrival Heart attack patients given percutaneous coronary intervention (PCI) within 90 minutes of arrival Average (median) time patients spent in the emergency department, before they were admitted Average time patients spent in the emergency department before being sent home Average time patients spent in the emergency department before they were seen by a health care professional Percentage of patients who came to the emergency department with stroke symptoms who received brain scan results within 45 minutes of arrival Heart failure patients given discharge instructions Pneumonia patients given the most appropriate initial antibiotic Outpatients having surgery who got an antibiotic at the right time - within one hour before surgery Outpatients having surgery who got the right kind of antibiotic Surgery patients who were given the right kind of antibiotic to help prevent infection Patients having surgery who were actively warmed in the operating room or whose body temperature was near normal by the end of surgery Patients who got treatment at the right time (within 24 hours before or after their surgery) to help prevent blood clots after certain types of surgery As the above list demonstrates, there is a vast amount of value to be uncovered by using this information. HPL Claims Frequency Model Hospital Compare Database - Sample Timely and Effective Care Variable Figure 1 Indicated Relative Frequency 1.210 0 Figure 1 is just one variable from the timely and effective care category that proved to be a significant source of differentiation between high and low risks for HPL claims. Our research shows that hospitals, which had lower scores regarding outpatients receiving the right kind of antibiotic during surgery, demonstrated 21% higher frequencies. Outpatients having surgery who got the right kind of antibiotic Figure 2 HPL Claims Primary Layer Severity Model Hospital Compare Database - Sample Timely and Effective Care Variable Our research also shows that there was value to be gained by using severity models to analyze the Hospital Compare database. In Figure 2, we show a variable relating to the timely and effective care of patients. In this case, a lower quality care score was indicative of increased severity in the primary layer. Indicated Relative Severity 0 1.058 Heart Attack Patients Given Beta Blocker at Discharge 3 PINNACLE MONOGRAPH
HPL Claims Excess Layer Frequency Model Hospital Compare Database - Sample Timely and Effective Care Variable Figure 3 1.290 Indicated Relative Frequency 0 Finally, we show an example from the Excess Layer Frequency model (Figure 3). Though there is less data available at higher layers of loss, we found that there is still valuable information to be gained from the analysis. Figure 3 shows that for heart attack patients given PCI within 90 minutes of arrival, a lower score was an indicator of increased large loss frequency. This specific variable indicated nearly a 30% increase. Readmissions, Complications and Deaths In addition to the quality of care measures, there are several indicators of potential issues due to readmissions, Heart Attack Patients Given PCI Within 90 Minutes Of Arrival complications, and deaths. These variables were also found to be significant sources of information pertaining to HPL claims. They indicate potential errors that were made during initial patient contact and identify the rate at which these incidents are occurring at individual hospitals. Mismatched blood types, for instance, indicate that there was an error made at some point during care of the patient and are the type of error that could potentially lead to significant HPL claims. In our analysis, we found several of these variables to be indicative of future HPL claims. This category also contained several never events. Death rate for heart attack patients Death rate for heart failure patients Death rate for pneumonia patients Rate of readmission for heart attack patients Rate of readmission for heart failure patients Rate of readmission for pneumonia patients Rate of readmission after discharge from hospital (hospital-wide) Collapsed lung due to medical treatment Serious blood clots after surgery Accidental cuts and tears from medical treatment Deaths among patients with serious treatable complications after surgery Objects accidentally left in the body after surgery Mismatched blood types Falls and injuries Blood infection from a catheter in a large vein HPL Claims Primary Layer Frequency Model Hospital Compare Database Sample Readmissions, Complications, and Death Variable Figure 4 Indicated Relative Frequency 1.203 0 The results from the frequency model for a variable from the database related to Readmissions, Complication, and Death rates are shown in Figure 4. It shows that higher heart attack death rates indicate a significant increase in the frequency of HPL claims. Heart Attack Death (Mortality) Rates PINNACLE MONOGRAPH 4
Use of Medical Imaging This category relates to how a given hospital uses medical imaging tests on their patients. The variables indicate the frequency of these types of tests within a given hospital. Outpatients with low back pain who had an MRI without trying recommended treatments first, such as physical therapy Outpatients who got cardiac imaging stress tests before low-risk outpatient surgery Outpatients with brain CT scans who got a sinus CT scan at the same time Survey of Patients Experiences The surveys of patients experience opens up a wealth of information regarding how patients view their own experiences. For HPL claims, this can be very valuable information. When the communication between the medical staff and the patients is frequent and of high quality, there is a chance for fewer possibilities of errors to occur in patient care. In addition, understanding how patients view certain hospitals can be an indicator of how well that hospital is performing in relation to other hospitals. These concepts were used in the models and we found that, in general, better communication and happier patients lead to lower frequency and severity for HPL claims. Patients who reported that their nurses Always communicated well Patients who reported that their doctors Always communicated well Patients who reported that they Always received help as soon as they wanted Patients who reported that their pain was Always well controlled Patients who reported that staff Always explained about medicines before giving it to them Patients who reported that their room and bathroom were Always clean Patients at each hospital who reported that YES, they were given information about what to do during their recovery at home Patients who reported YES, they would definitely recommend the hospital Figure 5 HPL Claims Primary Layer Frequency Model Hospital Compare Database - Sample Survey of Patient's Experiences Variable 1.260 Figure 5 demonstrates the value of an individual variable from the primary layer frequency model. In the model, this Survey of Patients Experiences variable indicates a difference of more than 20% in the frequency of HPL claims between hospitals with lower scores and higher scores. Within the context of the frequency model, this variable indicated that hospitals with a lower percentage of patients with poor hospital experiences were significantly less likely to have an HPL claim. Indicated Relative Frequency 0 Percent of patients who responded they would not recommend the hospital 5 PINNACLE MONOGRAPH
HPL Claims Primary Layer Severity Model Hospital Compare Database Figure 6 Indicated Relative Severity General Hospital Information A final category of information pertaining to the hospitals relates to general information regarding the location and features of the hospitals (Figure 6). The ownership of the hospital is an interesting variable as we were able to investigate distinctions between government-owned and privately-owned hospitals. Though we do not show the scale of the differences, it can be seen that the different types of hospital ownership lead to different levels of severity for HPL claims. Some of these differences may be intuitive, but with the aid of the Hospital Compare database, we can quantify the amount of the differences. Overall Findings Hospital Ownership Once the various models were completed, we developed overall risk scores at the individual hospital level. At this level, the relative HPL experience between hospitals can be used for pricing or risk management benchmarking. To demonstrate how the results might be used, we have compiled an example of six hospitals across the United States. The intention is to demonstrate the value of being able to understand the underlying risks that can arise from difference between hospitals. For each hospital, we indicated the relative level of risk for each of the variables in the model. Red cells indicate the hospital was in the high risk category of a given variable. Green indicates that the hospital was in the lower risk category of a given variable. The black cells are variables that were not included in this particular model. For this exercise we focused on primary layer frequency. Primary Layer Frequency Model Hospital Compare Variable Hospital Ownership Variable 1 Medicare Payment Variable 1 Survey of Patients Experiences Variable 1 Survey of Patients Experiences Variable 2 Survey of Patients Experiences Variable 3 Readmissions, Complications, and Deaths Variable 1 Readmissions, Complications, and Deaths Variable 2 Timely and Effective Care Variable 1 Timely and Effective Care Variable 2 Timely and Effective Care Variable 3 Timely and Effective Care Variable 4 Timely and Effective Care Variable 5 Timely and Effective Care Variable 6 Timely and Effective Care Variable 7 Timely and Effective Care Variable 8 Timely and Effective Care Variable 9 Timely and Effective Care Variable 10 Hospital A B C D E F City X X Y Y Z Z State AA AA BB BB CC CC Private Church Government - State Variable Level Church Private Church Indicated Relativity 1.18 0.98 1.78 1.11 0.77 1.44 The above chart identifies the relationship of the six chosen hospitals within the context of the primary layer frequency model. Looking across each variable, it can be seen which hospitals performed better or worse in relation to that given variable. Looking down each column, it can be seen how often the hospital is performing better or worse across the multiple categories. The Indicated Relativity row demonstrates the total frequency relativity indicated by the model. In general, the more red flags a hospital has, the greater the indicated frequency relativity. For example, Hospital F has more red flags than Hospital E, which corresponds to almost twice the indicated frequency per exposure of Hospital E. However, due to the relative frequency for State Government run facilities, Hospital C has the highest indicated relativity with the lowest number of red flags. Not shown in the chart are the relative weights that each variable contributes to the total relativities. It would be pertinent for an underwriter or risk manager to examine the relativity of each variable and to understand which indicator of poor risk is driving the experience for a hospital. PINNACLE MONOGRAPH 6
Figure 7 Predicted versus Actual All Layer Validation Frequency Comparison 6.00 10.3% Model Performance As shown in Figure 7, the frequency models identified a spread of risk of approximately 5.50 from the highest to lowest frequency. Thus, our predictions of frequency were 5.50 times greater in the group of hospitals with the highest frequency as compared to the group with the lowest frequency. In the groups between high and low we see a fairly consistent progression of increasing relativities which track fairly well to actual loss relativities. We also observed similar results in the overall severity model, where the spread from high to low was roughly 3.5 to 1.0. Frequency Relativity 5.50 5.00 4.50 4.00 3.50 3.00 2.50 2.00 1 2 3 4 5 6 7 8 9 10 10.2% 10.1% 10.0% 9.9% 9.8% 9.7% 9.6% Percentage of Total Exposure Exposure Groups (Ranked Ordered by Predicted Frequency) Exposure Predicted Frequency Actual Frequency Predicted versus Actual All Layer Validation Loss Cost Comparison Figure 8 2.50 10.2% Loss Cost Relativity 2.25 2.00 1.75 10.1% 10.1% 10.0% 10.0% 9.9% 9.9% 9.8% Percentage of Total Exposure Combining the results of our frequency model and severity model, we were able to determine the average loss cost per base class equivalent exposure (i.e., occupied acute care bed). Going through a similar validation process as frequency and severity produces the following graph (Figure 8). The spread of the loss costs is approximately 2.2 to 1.0. In other words, HPL insurance costs per acute care bed for the worst performing hospitals were more than double the costs for the best performing hospitals. 1 2 3 4 5 6 7 8 9 10 Exposure Groups (Rank Ordered by Predicted Loss Cost) Exposure Predicted Loss Cost Actual Loss Cost 9.8% Conclusion The megatrends driving health care risk analytics are here to stay. So too are the fundamental changes in how health care is being delivered in the United States particularly outcome-based compensation and the decline of the sole practitioner physician in favor of physician groups and hospitals directly employing physicians. In this dynamic environment, hospital risk managers and the HPL insurance risk bearing entities (insurers, reinsurers, captives and RRGs) need tools that will help them identify and quantify risk factors impacting the frequency and severity of negative patient outcomes and HPL claims. In this report, we have introduced the free Hospital Compare database available at www.medicare.gov. Our analysis has demonstrated that the Hospital Compare database, when combined with historical HPL exposure and claims experience can be a powerful tool to identify and quantify controllable characteristics that impact HPL claims frequencies and severities. The tool shows that certain characteristics of quality care in hospitals can lead to higher or lower loss potentials for hospitals. Beyond the scope of this research, we believe that the Hospital Compare database and other similar databases are going to provide significant value in the HPL insurance industry. We believe that as the database matures, there will be the potential for additional risk identifiers beyond those which we have already identified. As more years of data are collected, we will also be able to examine trends over time in specific characteristics and determine how improvements in a key metric help identify changing risks. Also, as ACA becomes the standard for health care in the United States, we see the potential for additional mandatory data reporting requirements, which could then be considered for analysis. In total, we believe the Hospital Compare database already provides significant risk management value; future improvements and analysis will only lead to more discoveries. 7 PINNACLE MONOGRAPH
PRESORTED FIRST-CLASS MAIL U.S. POSTAGE PAID BLOOMINGTON, IL PERMIT NO. 111 2817 Reed Road, Suite 2, Bloomington, IL 61704 Bloomington I Atlanta I Chicago I Des Moines I Indianapolis I San Francisco Commitment Beyond Numbers October 2013 Monograph Using the Hospital Medicare Licensee pinnacleactuaries.com Driving better business decisions means seeing the road ahead clearly. Lee W. Knepler, ACAS, MAAA, Robert J. Walling, III, FCAS, MAAA, CERA KEY POINT Health care risk analytics is at the confluence of two megatrends of the last decade health care risk and risk analytics. Analytics can be used to not only predict health care risks, but to reduce or avoid them. Risk analytics - statistical tools that use Big Data to reduce, mitigate or predict risk - have impacted almost every aspect of our lives. In this research report we have analyzed data from the Hospital Compare database to examine the impact certain quality of care characteristics have on hospital HPL costs. To discover what we mean when we say Commitment Beyond Numbers, visit us at pinnacleactuaries.com.