Today physicians are increasingly dissatisfied with the practice of medicine, especially primary care providers. According to Merritt Hawkins’ 2014 Survey of America’s Physicians, 44% of doctors plan to cut back on patients seen, retire, work part-time, or seek a non-clinical jobs. This comes at exactly the wrong time. It is predicted that we will have a critical shortage of 46,000 to 90,000 doctors relative to the increasing needs of an aging population.
There are many reasons for this exodus. A common reason is that physicians are being asked to do more and more activities not directly tied to patient care. Not only does the physician need to chart the patient visit, but she must select diagnosis and procedure codes for billing, address health maintenance tasks, fill out referrals and prior authorization requests, reconcile patient medication use, reply to social service forms, and so on.
Many primary care physician feel these days that talking with and caring for the patient is a side job. In fact, in many medical offices these days, the patient talks to the back of the physician as she is staring at a computer monitor, clicking a mouse, and typing on a keyboard—not exactly conducive to establishing therapeutic trust.
With the increasing adoption of electronic medical records, there have been studies which show decreasing productivity among physicians—one study found that ER doctors spend 43 percent of their time on data entry—exactly the opposite of what was promised.
Among the non-clinical activities which physicians are mandated to perform, an important one that suffers is the proper selection of the codes which indicate pertinent patient diagnoses and the treatments provided during an encounter. These codes are used for proper payment and are also increasingly used for data analytics related to risk prediction, quality of care, practice patterns, and resource allocation.
There are tens of thousands of diagnosis and procedure codes from which to choose, and the code set is growing with the adoption this year of a new set of diagnoses and procedures, otherwise known as ICD-10. In the older diagnosis set, ICD-9, there are 40 different codes for diabetes, which includes different manifestations of the disease such as kidney disease. In ICD-10 there are over 110 codes for diabetes. Negotiating this transition is a bewildering activity for even trained coders.
In an era of payment-for-value, rather than payment-for-services, these codes become an important set of data. The activity of coding should not be left to the overworked physician with typically little to no formal training in coding. Improper diagnosis or procedure code selection following a patient visit in the clinic or hospital can result in inappropriate payment (too much or too little), inaccurate risk or quality performance measure results, or poor targeting for proactive management of costly patients.
One study undertaken in 2015 by a vendor which manually reviewed 100,000 charts from practices across 11 states found that 28 percent of the documented conditions were not coded on the billing claims submitted to the health plans. These codes were not essential for the physician to get paid for the service, but they are useful for other activities. This phenomenon, known as under-coding, causes health plans and healthcare systems to falsely conclude that their populations are healthier than they actually are.
Rather than try and clean up the issues created by physician coding, it makes more sense to train physicians to document well, and leave the coding exercise to the experts.
A cottage industry of certified (professional) coders who rework the submitted codes has grown up around poor physician coding. By reading the clinical documentation, these coders decipher what should have been the optimal code to represent a diagnosis or treatment provided.
It should be coders, not physicians, selecting diagnosis and procedure codes. The only obstacle there is that there is much more work than trained coders, especially with the transition to ICD-10. My team at Apixio has developed a web application to support coding activities. Built upon the insights from analyzing more than 50 million patient documents, our solution provides highly accurate text mining of charts for chronic conditions which attribute to significant financial costs and patient morbidity. And coders can review two or three times more charts during a given period of time than they are able to do on their own.
We can and should lean on coders, working in conjunction with cognitive computing trained to read medical charts, to allow physicians and their office staff to spend less time doing taxing back-office work and more time treating patients. It will enhance physician satisfaction and elevate the long-term sustainability of the medical profession for many doctors.