Commercial insurers, government health plans, and self-insured employers have many opportunities to leverage big data and big data technologies. Medicare and some of the largest private health insurers have member populations and claims volumes that easily qualify as “big data” on their own. But even greater insights are made possible by combining data from multiple payers and increasing the number of claims to detect new patterns and establish correlations with strong statistical validity.
For some applications, HIPAA restrictions on data use must be applied because fully identifiable patient information is needed. For others, large de-identified databases can provide valuable insights into healthcare claim patterns that can enable better care management. Some notable repositories of de-identified longitudinal claims data include:
- Medicare claims from CMS
- Multi-payer claims from Truven Health Analytics
- Blue Cross claims from Blue Health Intelligence
Payers can use big data to improve care, including:
Reducing Fraud, Waste and Abuse
Reducing Fraud, Waste and Abuse
Effective programs to prevent deliberate fraud can have a huge impact on reducing the overall cost of care. The FBI estimates that healthcare fraud is an $80 billion a year problem in the US. Basic steps to ensure payment integrity do not require big data techniques and involve reviewing individual claims using pre-determined rules.
These steps include:
- Checking on eligibility of the patient and the provider
- Avoiding duplicate payments
- Reviewing diagnosis codes to ensure that the care was medically appropriate
Creative criminals are constantly finding new, hard-to-detect fraud schemes; big data techniques can be applied to large volumes of claims to reveal suspicious patterns to help avoid these abuses. One approach is to look for providers with unusually high volumes of claims for types of services that are often subject to abuse. Combining data from multiple payers and creating a consistent provider identifier across those claims streams can highlight problems that might not be obvious from just one health plan. In addition to recognizing unusually high volumes of billed services, analysis of big, combined databases can also highlight providers whose charges are consistently higher than others treating similar patients. High charges may reflect abuses such as improper coding.
These anti-fraud analytic techniques can identify suspicious patterns which then require further investigation to determine if criminal activity is really involved. As with medical diagnostic tests, the best analytic techniques are those that minimize both false negatives and false positives. False negatives will mean that fraud is undetected and healthcare funds are wasted. False positives will mean that investigators waste time reviewing cases that turn out to be legitimate.
Similar techniques can be used to identify questionable claims where the issue may not be criminal but where medical standards of care may not be followed. For example, some types of expensive claims require manual review of clinical notes to make sure that the care is medically appropriate. This type of work has traditionally been done by specially trained nurses. By combining optical character recognition, fuzzy logic and text analytics, insurers can review clinical notes in an automated process. The straight-forward cases with good documentation can be paid without manual review, freeing up the nurses to review only the more questionable or poorly documented cases.
For payers, the most productive care management system is one that intervenes early to prevent the need for more expensive care later on.
Big data analysis can help payers to:
- Identify what preventive measures really work to reduce costs
- Provide case management early in the progression of a medical problem
Analysis of the impacts of preventive care and therapeutic compliance on future health costs can help health plans focus their efforts.
Example: Diabetic populations have high healthcare costs, and there are several established clinical standards for care of diabetic patients. But on which patients and specific interventions should a payer focus its care management resources? One effort to answer this question was undertaken by a group of Blue Cross plans from several states that pooled their claims data for to create a predictive model.
The pooled claims included 4 million patients with a diagnosis of diabetes. The statistical study looked at over 120 potential predictors to see what risk factors and interventions had the biggest impact on the likelihood that a diabetic would be admitted to a hospital for acute care.
The analysis confirmed the value of ambulatory care services such as glucose and cholesterol screenings in reducing hospitalizations. It also highlighted the importance of specific risk factors that were predictive of future acute care costs. Patients in specific categories defined by race, age, comorbidities and prior utilization were shown to have a significantly higher risk of hospitalization.
With these insights from big data, health plans gained the information needed to stratify patients and focus case management resources on effective preventive care for the patients at highest risk. They also gain the information needed to design quality incentive programs for providers that focus on the types of preventive care that have the strongest evidence and the highest cost impact.
Early Case Management
Some medical conditions do not have the same level of consensus on diagnosis and treatment standards as diabetes, but, like diabetes, can be associated with high cost acute care events. Back pain is a fairly common condition that in some cases can result in expensive surgical treatments. Other patients find relief with alternative treatments that are less invasive, less risky and less expensive.
Some payers are starting to experiment with big data statistical techniques to help identify patients heading down the path towards surgery. For those patients, they can intervene early with case management and referrals to providers who may be able to treat the patient without surgery. But how to identify those patients in time? Map Reduce algorithms and nPath time series analysis can churn through huge volumes of claims history and isolate risk factors that predict the likelihood of back surgery months or years in the future. Such predictors could include pain medications, physical therapy or massage. With a predictive model to identify those patients at risk, health plan case managers can intervene early to guide the patient towards effective non-surgical treatment.
Healthcare payers, including the Centers for Medicaid and Medicare Services (CMS), are banking on informed, engaged consumers to play an increasingly important role in driving down healthcare costs. Big data is helping drive consumer engagement in several areas, especially in:
- Customer service
Payers hope information about the costs and quality of different providers will be an important driver of patient behavior in care selection. Patients are more sensitive to costs in general due to rising copayments and deductibles, but the lack of good, actionable price and quality information has limited the impact on the selection of hospitals and physicians.
Quality measurement is notoriously difficult, but big data can help. National claims databases can include providers from every state and every major city and can encompass a high enough volume of claims for common medical problems to make statistically meaningful comparisons of complication rates and other measures of provider quality. It is not clear, however, that just making quality data available to consumers really makes much of a difference in their choices. The publication on the CMS website of Hospital Compare has not had a big impact on consumers so far.
However, private insurers are using similar quality data as part of their criteria for forming provider networks. Insurers then provide financial incentives for plan members to use the preferred network providers.
Example: Plans use quality measures from pooled national databases to help select their centers of excellence for particular service lines. Tiered and narrow networks use varying levels of patient copays and deductibles to motivate patient behavior.
With large copay and high deductible responsibility, consumers may be quite interested in knowing the price of a service ahead of time. However, that information has rarely been available for inpatient hospital services. It’s become more common for hospitals to provide pre-service estimates as part of their effort to collect copayments upfront, but firm prices have been rare.
Reference pricing is a new approach to price transparency that depends heavily on big data and seems to have a real impact on consumer behavior. In California, the pension plan Calpers and the insurer WellPoint collaborated on a program in which they explicitly set a maximum hospital payment (reference price) for total hip and total knee replacements. The reference price was set after analysis of the wide range of variation in prices in historical claims.
Members received a letter notifying them which hospitals were below the $30,000 reference price and told they would have minimal out-of-pocket costs at those hospitals. Patients choosing a hospital with higher prices were responsible for the extra cost above $30,000.
The program resulted in a significant shift in volume to the selected hospitals in the program, and the plan saved money. There were additional cost savings during the year as some hospitals that were not initially in the program dropped their prices. Claims analysis showed that outcomes were the same or slightly better for patients at program hospitals. This experiment shows promise for price transparency having an impact patient’s choice of inpatient services – at least in markets where there are multiple hospitals competing for business.
Satisfied customers are increasingly and directly important to the financial health of commercial insurers.
For insurers offering Medicare Advantage plans, CMS payments are higher if member satisfaction is high. Satisfaction surveys, retention rates and member complaints all factor into CMS’ Star rating that is calculated for every Medicare Advantage plan. Higher-rated plans receive bonus payments, and, at least once per year, dissatisfied members have the option to switch to another plan. Departing members result in a loss of revenue and – even if they can be replaced with new members – the costs to recruit and onboard new members are quite significant.
Some Medicare Advantage plans are starting to use big data analysis to help identify and address customer service problems in order to maintain high levels of satisfaction. The first step involves building a predictive model to identify customer service events that significantly increase the chance that a member will file a complaint or switch out of the plan. To build a comprehensive model, data from a wide variety of sources must be combined.
Important data include:
- Call center logs
- Web site traffic
- Mailings of “explanation of benefits”
- Claims processing delays
- Billing errors
Map Reduce algorithms and nPath time series analysis can determine which specific types of events are correlated with complaints and departures in later months. With this predictive model in place, the health plan can work to reduce the frequency of those problems, quickly recognize them when they occur, and take rapid remediable actions to address the customer issue.
HIMSS Clinical & Business Intelligence Primer
Module 1: What is Clinical Business Intelligence (CBI)?
Module 2: The Urgent Need for Clinical Business Intelligence
Module 3: Types of Clinical Analytics
Module 4: Essential Technologies for Clinical Analytics
Module 5: Organizing for Success in Clinical Analytics
Module 6: Future Directions for Clinical Business Intelligence