First and foremost, the sector must expand the focus from reserving and underwriting to gathering better data. Without better data, and the ability to practically use this data, companies will continue to struggle to identify claims that could benefit from additional oversight throughout the life of the claim.
As currently constituted, most of these systems are able to capture, for example, descriptions of the injury or illness that spurred the claim. But much of this information is vague in nature. Most claims systems capture the primary body part injured, but this does not get to the clinical severity of the injury.
The most common injury type for almost all organizations, for example, is lower back injuries. The description "back injury" is too generic to really target underlying risk factors, but if the system is able to capture the primary diagnosis codes identified through medical billing, companies could better distill the real clinical severity of the injured worker. Better precision could improve the process drastically.
For instance, including medical codes 722.7 (for intervertebral disc disorder with myelopathy) or 846.0 (for a sprained/strained lumbosacral joint)would help those analyzing the information get a clearer picture of the claim. Most professionals experienced with these two diagnoses agree that an injured worker with a disc disorder will most likely have a more complex claim than one with a general strain of the back.
However, even though we know that a disc disorder can be associated with additional risk, not all injured workers with this diagnosis will respond to treatment the same way and often will require different levels of oversight. What if one injured office worker is a 65-year-old male with diabetes who travels 50 miles to work everyday? The claim professional may implement a different strategy than she would for an otherwise-healthy, 30-year-old male who commutes five miles to his office job. A complex or lengthy commute could be a huge barrier, discouraging the injured worker from getting back to work sooner.
Unfortunately, due to the current means of data capture, many of these factors remain hidden. They lie only within the notes of the injured worker's file, which cannot be easily accessed by models, rather than stored in discreet data fields that predictive models can incorporate into the overall risk assessment.
Generally, today's predictive models include the following fields: age, gender, cause of the injury, nature of the injury, work status and job classification. To create better predictive modeling tools, organizations should increase the categories of data they capture, perhaps including: prescribed medications (specifically narcotics), socioeconomic factors (such as education level), psycho-social factors (such as job satisfaction) and distance to the primary work site.
Much of this information is included in notes by the claims handler, but if it is not in its own data field, most predictive models will not be able to find it. A number of predictive modeling tools can utilize the free-form text data that is found in notes, but they are limited in their ability when compared to those that use structured data.
Data Warehouse Development
There is also other data that can help the claims process. And to leverage their investments in predictive modeling, many organizations are attempting to bring all of the data associated with an injured worker into centralized data warehouses. This includes, but is not limited to, medical billing data from the bill review systems, pharmacy transactions from the pharmacy benefit management program and nurse activities from their case management partners.
Looking at even this abbreviated list of data sources can quickly overwhelm even the strongest analytic supporters within an organization. The process of gathering the information from multiple partners and mapping it into a data warehouse is time- and resource-intensive. Some organizations might consider it nearly impossible to achieve the ultimate end goal of a quality data set.
But the value of this information is definitely worth the investment. While the actual financial payoff will vary from one organization to another, there is no doubt that a data warehouse of this scale can promote better clinical case management and lead to reduced medical and indemnity costs. Most importantly for everyone involved, it will also get employees back to work more quickly.
Think about the number of areas that could use an integrated view of the medical transactions alone. Given our historic medical cost trends, the ability to use this detailed treatment information as part of a risk modeling capability is tremendous.
When one considers the example of the 30-year-old with a back diagnosis, identifying him as high risk is unlikely given the data elements typically available. If we include medical utilization patterns, however, our predictive model may make a more accurate forecast. If the model identified, say, narcotic use, with increasing strength, continuing for 30 days after the injury date, that would most likely be a predictor of a higher-risk injury given the need for pain medication. Without a consolidated data source, the predictive model is blind to this new, invaluable information.
Not all organizations have the same appetite to develop and implement these tools. Many companies prefer to leverage the human element and would rather trust the experienced professional managing their claims to identify risk. That is understandable. The fact that there is more art than science to identifying risk is an oft-heard refrain. While many will attest that much of what can be identified by a predictive model could be identified by a good claim professional, it is becoming more evident that these human resources are generally stretched thin with endless tasks. This means leveraging a predictive model could help ease their burden and ensure that every factor of every claim is being weighed.
It is true that the best claims professionals can see any claim better than even the best model. But the best models can see every claim. That ability, however, is only as strong as the data it has to work with.