How to Get Really Good Analytics

Many in Workers’ Compensation are now turning to analytics, searching for the last best solution to controlling costs. Analytics will go a long way to meet that expectation if applied correctly. However, the components of good analytics may be elusive for many. The only way to produce really good analytics is to build them using really good data.

Good data will produce useful analytics
To say the prerequisite for analytics is good data is a bit simplistic because many factors are involved. Yet, good analytics does require good data before anything else. The volume of data is not as important as quality and content of the data. Stated even more strongly, only extraordinarily pristine and comprehensive data can result in uniquely useful analytics. The opposite is also true. Bad data can never result in good analytics. So, let’s discuss what makes for good data.

Our recent article, “WC Analytics Can't Live on Bill Review Data Alone” suggests that while bill review data is good, it is not enough. It’s scope is too narrow. The article focuses on why good analytics cannot be built on bill review data alone, as some propose. But that is not the only problem. There are many more conditions, omissions, and misapplications of data in the industry that limit the usefulness of analysis. Moreover, most are straightforward, simple issues. A few of them are the following.

System design effects data quality
One source of bad data is poor system design. Omitted data fields and awkward or illogical flow confuse users or force them to enter misleading data. Unfortunately, many users are clever at “beating the system”, thereby creating unusable data. But an even bigger problem is data systems that simply do not contain important data fields. One critical data field that is often missing is physician specialty.

Systems frequently omit physician specialty
To be fair, physician specialty has not been an important data element until recently. Now that medical costs amount to 60% of claim costs, deeper analysis of cost drivers is of considerable importance. Medical doctors exert a pronounced effect on the outcome of claims and their performance should be evaluated and rated. However, it doesn’t seem fair or logical to compare the performance of an emergency department physician with that of a neurologist. Comparing a psychiatrist with an orthopedic surgeon is just as unreasonable. Yet, when the data lumps all medical doctors into the same category, more precise analysis is not possible. Adding the one data element of physician specialty, analysis can rationally target costs in claims.

Physician NPI numbers are critical
Another critical data element frequently omitted from data sets is a physician unique identifier. Reliance on tax ID alone is no longer acceptable. In the past, the tax ID was the only data element of interest, in order to accurately pay the bill. But for good provider performance analytics, accurately identifying the individual provider is vital. Provider performance cannot be analyzed unless individuals can be differentiated in the data.

It’s true; some treating providers seek ways to obfuscate their individual identity by using multiple tax ID numbers or conceal their identify behind organizational or facility tax ID numbers. A few are even using multiple NPI numbers, and not all of them registered.

NPI (National Provider Identifier) is a national registry of individual medical providers with unique numbers for individuals. Medical providers, but especially medical doctors, should register and obtain an NPI umber. In fact, to be reimbursed for group health medical services, an NPI number is required. Unfortunately, NPI’s are not required for Workers’ Compensation reimbursement.

Taken a step further, claim systems should insist upon and utilize the NPI number to identify individual providers as a condition of payment, similar to group health. The practice would prevent fraudulent gaming of the system on that level. Requiring and implementing the NPI number is a simple step that would achieve powerful analytic results because individuals would be recognized accurately.

Good data is a management imperativeUltimately good data is a management function. Systems will not improve until organizational leadership mandates it. Business management must share this responsibility with IT, not simply delegate the responsibility to those who may not grasp the business impact of poor or missing data.

Additionally, business leadership should aggressively establish data entry accuracy accountabilities. Data entry persons must be held accountable for data entry errors that effect the system and the organization. For instance, multiple records in the data for the same person or entity significantly diminish data quality and the ability to product good analytics.

Too often, the data entry person creates a new record for a vendor, medical provider, or other entity rather than search out the appropriate record already in the system. Creating a new record automatically creates a unique record and results in duplicate records for the same person or entity, each with different identifying numbers. The system regards each record as different and unique, thereby rendering the system inaccurate. Basing analytics on such data results in misinformation in the organization. Insisting upon accuracy in this regard is a management imperative.

These are but a few of the data issues facing developers of analytics in Workers’ Compensation. Suffice it to say, poor quality data directly impacts the accuracy of information, and, therefore, the quality of products, services and outcomes of the organization. Conversely, those who insist on good and accurate data will enjoy the benefits of good analytics.

Learn more about Workers' Comp Analytics: MedMetrics Blogs