How to improve reporting of quality measures
We live in an era where healthcare is measured and rated just like everything else we purchase and consume. The problem is, rating healthcare is not as straightforward as measuring a consumer product or service.
Not all patients or methods of healthcare delivery are the same. Not all physicians are above average. And gathering, curating and interpreting the data needed to shed light on good providers is not easy.
A recent study estimated that it costs providers over $15 billion each year to report upon quality measures. Even though there are many attempts to rate the quality of physicians and hospitals, the accuracy and usefulness of the results is poor. We know more about the cars we drive than the heart surgeon who is about to cut our chest open.
There’s a reason for this. With many purchases, such as cars, we plan purchases in advance, and thus are able to research and make informed decisions. But, unlike cars, not all healthcare can be purchased ahead of time. When you are in the ambulance driving to the ER with chest pain, you are not going to look up quality scores to decide which hospital to choose. However, for healthcare services that are considered “preference-sensitive,” you have the opportunity to choose among different providers and sites of care—in those cases, quality ratings do and should matter.
Over the past three decades, there have been many efforts to rate providers’ quality of care. Federal bodies (such as the National Committee on Quality Assurance) and medical specialty societies (for example, the American College of Cardiology) have taken “best practices” based upon evidence of what works and translated them to formulas (“measures”) to rate provider quality. Examples of such measures would be the percentage of diabetic patients who have received a test for average blood glucose (hemoglobin A1c) every six months. The scores compiled for these measures have been used to determine performance-based bonuses and quality report cards.
To reduce the burden on providers, current measures have been compiled from easily available data created for billing and administrative purposes. This data is commonly obtained from the claims used by providers to get paid for services from health insurance companies. The data on the claims is structured into codes for diagnoses, procedures, treatments and physician visits using commonly accepted code sets, such as the International Classification of Diseases, or ICD).
We can obtain these structured data from an electronic health record in which the physician office will translate clinical activities into billing codes to receive payment from health insurance companies. What is not present in this data is the clinical nuance and the context that describes individual health and healthcare. What were the aspects of your diabetes (not just is it present)? Is your diabetes poorly controlled and associated with worsening kidney and heart function? Do you have a family history of disease? What other types of care are you receiving or have received?
The reliance on structured billing data for computing quality metrics limits the applicability of the ratings. With only billing data, we can determine whether best processes have been followed, not whether the outcome was optimal. Was a vaccination provided to kids? Was a blood test done for a diabetic to monitor blood sugar levels? Was a mammogram performed to detect breast cancer?
While it is important to ensure that these processes are followed for basic prevention or secondary prevention of disease, these process-oriented measures do not indicate whether outcomes were good or not. For example, a physician might have ordered a blood sugar level in a diabetic but later failed to prescribe the best treatment regimen for the patient. To know that, we need to have richer detail about the patient’s clinical care, and that can best be determined from an analysis of the clinical chart notes.
With the ability to use computers to read a patient chart and assemble a care profile (a phenotype), we can take into account individual factors and more completely capture treatments provided. And we don’t have to lean on overworked physician practices to manually assemble these data from reading charts.
With patient detail from the chart, we can compute an “apples to apples” comparison between physicians and hospitals and even patients. An academic hospital might appear to have a 30-day mortality rate after coronary heart surgery that is higher than a community hospital. This is not quite what might be expected. After considering that the patients who elect to undergo heart surgery at the academic medical center are much sicker than those who undergo surgery at the community hospital, it is the academic medical center that appears to have better outcomes than the community hospital when the underlying patient illness is taken into account.
Armed with the right kind of information, consumers can make choices according to the best value. And we can get much better insights into variations among providers. But for this to happen, we need to make sense of the 80 percent of a typical patient’s clinical information that is written in your medical record. Accessing the riches of this data, and making it useable, will enable strong quality ratings that change the delivery and consumption of healthcare as we know it.