Through our partnership with Risk & Insurance magazine and other media opportunities, we have written the following informative article.
Answers to Your Most Common Questions about Predictive Analytics in Workers’ Comp
Predictive analytics models have a number of cost-saving applications in the world of workers’ compensation prescription management.
These models are used operationally in helping to identify patients for home-delivery fulfillment programs which are less expensive than retail, saving money on the cost of prescriptions. Algorithms assist in routing claims to the most appropriate case handler based on claim complexity or personnel experience, which increases operational efficiency.
OWCA Chief Clinical Officer
Perhaps more importantly, “predictive analytics help to identify problems before they arise,” said Tron Emptage, R.Ph, MA, Chief Clinical Officer, Optum Workers’ Compensation and Auto No-Fault.
For example, a predictive model can raise red flags after detecting unusual prescribing patterns of a physician or inconsistencies in a claimant’s typical refill pattern. Alerting claims professionals to this data allows communication with both patient and provider, proactively mitigating future consequences like extended claim duration and over-utilization, which could lead to adverse drug interactions or addiction.
But for insights delivered by analytical tools to be effective, they must be paired with clinical expertise. Predictive analytics platforms can show potential paths forward on a claim and their likely outcomes, but they can’t make decisions alone. And, according to the experts at Optum, nor should they.
“Using analytics in combination with clinicians is how we effect change,” said Joe Anderson, Director of Analytics and Data Science. “Clinical expertise is an integral, not supplementary, component of leveraging data-driven insights to produce better outcomes both medically and financially.”
This may leave workers’ compensation claim payers wondering how best to integrate predictive analytics in their operations to generate the best outcomes from both clinical and cost perspectives. Here, Emptage and Anderson answer the seven most common questions companies ask about predictive analytics:
1. What is “predictive analytics” and how does that differ from other analytical tools?
There are no technical or scientific definitions for terms like predictive analytics, artificial intelligence, or machine learning, and the distinctions between them are fluid as technology continues to evolve.
“The simplest way to define predictive analytics is using an algorithm to make a decision. Data scientists will see fine distinctions between different platforms and tools, but the differences ultimately come down not to the technology, but how you use it,” Anderson said.
The key feature of predictive analytics is that it is forward-looking, meaning it evaluates several alternative decisions and projects their potential outcomes to aid in real-time decision making.
Traditional data analytics platforms, on the other hand, examine past data to identify trends and patterns. They are used for observation and learning, while predictive platforms are used to drive action.
2. What are the benefits of a forward-looking platform versus one that examines the past?
A traditional data analytics platform will help users spot correlations between different trends and make determinations about their common drivers. A particular medication, for example, may correlate with longer claim duration or higher costs, and one specific physician may be the source of those prescriptions.
The insights generated by traditional data analytics, however, are unchangeable. They provide a retrospective look at what’s already happened, but don’t always point to a way forward. Predictive analytics algorithms suggest immediate actions that can proactively alter the course of a claim.
“Predictive algorithms project the costs that an injured worker will generate in the future, and show what elements are driving that trend. In this case, there’s a more immediate judgment call to be made,” Anderson said.
“For example, if $50 was spent on a prescription last month, that number is what it is,” Emptage added. “A predictive model shows us how to spend less in the future. It’s not prescriptive on what decisions to make, but will help highlight the options and potential outcomes of those options.”
3. What data is most important for predictive analytics?
Identifying potential problems before they arise requires pulling together disparate data to create a holistic picture of a patient.
“Pharmacy data is most important,” Emptage said. “This includes the prescription data itself and the prescriber writing it, and whether the patient is seeing any other providers also writing prescriptions, similar or different.”
The injured worker’s diagnosis factors largely in the overall severity of a claim and in a predictive model. Even with a simple injury, supplemental data like previous medical history and comorbidities may indicate increased risk of a complicated and extended recovery. Psychosocial variables, including a patient’s social support network and general outlook on their recovery, also are strong indicators of how the claim will progress.
“Personal demographics may have some predictive value, but that information isn’t always as useful as what they’re getting in terms of medical treatment,” Emptage said.
4. With so much data available, why is clinical expertise necessary?
Predictive models were first built based on dollars. They could predict claim cost, but treatment decisions are based on much more than just the price tag. Clinical experts look beyond dollar signs to help make decisions that will produce the best medical outcomes for patients while balancing payers’ need to contain costs.
“You may have all the data in the world, but we are dealing with human beings, and clinical experts make sure we keep the human element of care,” Anderson said.
Clinicians can also spot when a model’s output simply doesn’t make sense.
A predictive model could, for instance, raise a red flag over a spike in a claimant’s morphine equivalency dose. “It may have missed, though, that the patient just had surgery,” Anderson said. “A clinical expert will see that and make sure there’s not an overreaction.”
“Predictive analytics is both an art and a science,” Emptage said. “We have the data, the algorithms, the output from models, but those results may turn out false positive or false negatives. Clinical expertise helps to weed out the false results.”
5. What are false positive and false negatives?
As in the above example, a model may produce a false positive if it indicates a problem where there is none, either due to incomplete data or simply due to a wrong prediction. Dedicating time and resources to solve a problem that never existed or never materializes can ultimately drive up the cost of a claim.
False negative may have more serious consequences. This is when a problem goes unnoticed and unaddressed. For example, a claimant may be taking excessive dosages of addictive painkillers due to concurrent prescriptions from different prescribers — but the model won’t pick up on this risk if it doesn’t have a complete list of the claimant’s providers.
“False negatives end up being even more costly because it usually means more care and a longer recovery,” Emptage said. “We don’t want to miss something that turns into a creeping catastrophic claim.”
This again is where a clinician’s input could make a pivotal difference. A trained expert will be able to spot the warning signs of addictive behavior even when the data says otherwise.
6. What are some predictive analytics best practices?
Using predictive analytics effectively comes down to three things: data, talent, and leadership.
Maintaining a database of historical data is necessary to feed a predictive model and ensure accurate results. “You can’t build models to predict the future if you don’t have data on what’s happened in the past,” Anderson said.
A model’s success is likewise dependent on the ability of specialists to interpret its findings and improve the algorithm over time. The model’s output should be continually compared to historical trends to identify deviations, and the algorithm should be adjusted accordingly.
Finally, a strategic commitment at the leadership level is critical.
“There will always be new types of data and treatments, products, clinical guidelines and regulations evolve. You have to be nimble and responsive to changing environment, and willing to keep an open mind about acting on data-driven insights,” Emptage said.
7. How will predictive analytics evolve in the future?
Cloud computing and open-source software tools — which help models run faster — may come to workers’ comp sooner rather than later as companies compete more heavily on the strength of their analytics capabilities.
Tools for unstructured data are also going to be big,” Anderson said. Currently much of the data collected comes in various formats —images, handwritten documents, emails, voicemails, etc. Tools capable of capturing and organizing this information will reduce the chance of error due to incomplete data.
Artificial intelligence platforms, in combination with unstructured data tools, may also be able to make and execute simple decisions by automatically generating new documents and records.
Predictive Models Provide Value in Workers’ Comp
Optum is helping to evolve the use of predictive analytics in workers’ comp. Thanks to its unique experience across the health system, working with providers, health plans, employers, government agencies and life sciences organizations, Optum is fluent in all of the types of data its clients use.
Its own predictive analytics platform, OptumIQ, brings together this data with advanced analytics and clinical expertise to better evaluate risk and drive more effective decision-making.
“We pull in more than just pharmacy data to create dynamic and constantly displayed risk score that will help adjusters or risk managers see how the patient is doing in real time,” Emptage said. “The need for intervention can be identified earlier, and predictive analytics helps us know where potential risks are and even what our options may be.”
“We can uncover risk and guide action, which brings value to both our clients and injured workers,” Anderson said.