Right Image, Right Time

Together, these drivers have created a more than $100 billion industry projected to grow as high as $161 billion by 2008. And, while these high-tech tests account for only 25 percent of all imaging procedures, they are responsible for 75 percent of the inflationary trend. Yet, the real problem is that at least one-third of advanced imaging testing is inappropriate, with more than $16 billion spent on unnecessary radiology tests.

Clearly physicians’ orders drive utilization of high-tech imaging. However, contrary to long-standing industry belief, overutilization of high-tech imaging is not the fault of a few doctors ordering inappropriately all the time. In reality, most physicians order inappropriately some of the time. Therefore, the $16 billion question is when and why does a particular doctor order appropriately or inappropriately? Finding the answer and managing the behavior can mean huge savings.

How PM Works for RM
Long employed in industries from financial services to air traffic control, PM is relatively new to healthcare and even more so to radiology management (RM). The reason for this slow adoption is that predictive models have only recently evolved to the required level of sophistication for healthcare.

First generation models provided a simple form of risk adjustment using basic patient demographic and clinical data such as age, gender and diagnosis, which provided a limited set of variables with limited efficacy. Second generation models used a broader set of member data to characterize members’ conditions. This facilitated development of methodologies that could classify or group patients into risk categories such as Ambulatory Care Groups.

Although they simplified patient analysis and management, these groups were neither specific nor sensitive enough and often led to care that was unnecessary, ineffective or incomplete. Today’s third-generation models employ artificial intelligence (AI), with these models typically falling into one of two categories:

1. Conventional AI models. These models or algorithms, employ formalism and statistical analysis. They operate on the principle that defined inputs to a series of defined steps will follow a defined path to a defined output. At its most basic, a recipe is an example of an algorithm.

2. Computational AI models. These models employ ‘learning’ or iterative development. They use empirical data as an input and change their structures and output based on external or internal information that flows through the network. Neural models are examples of computational systems with very strong pattern recognition capabilities. Thanks to this increased sophistication, PM can now be used to model not only which members will get the most sick, but also which doctors will order which imaging tests to diagnose and treat them.

All RM cost systems employ a sentinel component that saves payers money and improves care by:

• Denying inappropriate (unnecessary) tests. Most systems map ICD9 diagnosis codes against CPT-4 procedure codes, denying them when the former doesn’t support the latter. For example, if a physician ordered a computerized axial tomography scan of the head or brain (CPT-4 code 70450) for a patient diagnosed with a broken ankle (ICD9 code 824.0), the test would be denied.

• Approving a more appropriate test. In this instance, the inappropriate case is denied, but a more appropriate and/or efficacious test is approved. Using the broken ankle example, the physician should initially perform a basic X-ray of the ankle (CPT-4 code 73600), instead of requesting a CT or MRI of the ankle/lower extremity.

• Redirect patients to less expensive delivery locations. Imaging procedures performed in a hospital setting generally cost 100 percent more than those done at a freestanding imaging center. Yet, in the majority of markets nationwide, 60 percent to 70 percent of imaging utilization occurs in a hospital setting. RM systems can automatically suggest alternative, less expensive options to whoever is scheduling the test. The sentinel effect helps payers save current dollars and improve extant care. However, adding a predictive model takes radiology management a step further by providing prescient logic.

Employing a neural PM in RM allows a plan to:

• Identify inappropriate studies more accurately.

With the best neural models, decisions are based not only on CPT-4/ICD9 codes, but also such other factors as a particular patient’s demographics, the physician’s specialty and the physician’s past ordering habits. These additional variables do a more thorough job of identifying both potentially inappropriate and appropriate orders, thereby streamlining and improving the process. In most cases, the plan’s costs go down while quality goes up. In addition, physician satisfaction increases if the RM program rewards consistently appropriate ordering with fast approvals.

• Base approvals on ‘real’ data. Neural models can use experiential data to make predictions and they can analyze more variables. As variables change, the model adapts and provides real-world information. This ‘learned,’ more comprehensive knowledge gives plans a more accurate and comprehensive information base upon which to perform business planning.

• Manage by exception. The smarter the model, the more accurate decisions it can make to approve or deny up front. Only the questionable orders are forwarded for clinical review, or in some cases, review by physician medical directors. This provides the ability to apply valuable clinical decision support resources to the orders and ordering physicians who need them most.

• Adapt to market changes. Having a PM in place allows a plan to adjust the intensity of its utilization management program, thereby narrowing or widening its focus based on plan and market needs.

PM Today and Tomorrow
The advantages of PM for RM are many. However, despite their potential to actually improve the quality of care while reducing costs, the use of neural models for radiology management conjures images of provider profiling and denial of care. Clearly, radiology management with a prior authorization component leads to denied orders. Overutilization of high-tech radiology increases the cost of plans and reduces the quality of care, but proper utilization optimizes both costs and care.

But physicians typically don’t intentionally order inappropriately. The reality is that the changing pace of technology coupled with the increasing demands on a physician’s time make it almost impossible to stay abreast of what’s available and how best to use it. Physicians are human, and the most successful PM deployments for radiology management programs account for that by retaining the human element as part of the process.

Even the most evolved neural models cannot replace the critical thinking of experienced clinicians. Predictive models can successfully guide clinical practice when the variables are clearly defined and demonstrated. Fortunately, this is the case most of the time. However, when there are questions, the most successful radiology management programs forward them to real clinicians—nurses and physicians—to make the final decision about appropriateness. The best RM programs then facilitate peer-to-peer review of cases as needed. The result is that physicians have access to the evidence-based clinical decision support they need to provide the best in patient care.

APRIL 2007