Blind, Irresponsible Prescribing in the Age of AI?
The days of flipping through the PDR are long past as technology has outpaced referring to this age-old drug book now that computer programs can provide guidance.
Over the past decade, prescribing information has grown more and more complex, more and more dense and more difficult to negotiate. — Andrew von Eschenbach
“Take one and call me in the morning” is something far too many patients heard in past decades and, some, still hear it today. It is not only unacceptable, it is questionable medicine, IMHO. As their years in medical school fade into the distance, some healthcare providers sit by the side of the road with antiquated equipment and outdated thinking.
We live in an age of technology advancing at almost lightning-fast speed with developments unimaginable a few years ago. Knowing the difference between an agonist and an antagonist isn’t sufficient. Yet, in too many healthcare areas, doing what they’ve always done in the past continues to be the norm.
CMEs are supposed to keep practitioners up-to-date on medicine. I’ve been to a few presentations where the attendees didn’t bother to listen, texting on their phones instead. Their interests were the sign-in sheet, the coffee, Danish, and getting out as quickly as possible. Some signed in and left.
Not all CME presentations are medical advances, but the promotion of pharmaceuticals. Presenters may be paid thousands for their performance at these events. One has to wonder how this affects prescribing practices afterward.
Biochemistry Meets AI and Prescribing
The last hardcover edition of the Physicians’ Desk Reference (PDR) was published in 2017. It is now available free online as well as a free mobile app.
But it is a reference, although the software does provide information on whether medications can safely be combined. But what if the patient is already taking a variety of more than five medications and supplements? The task of AI here is to facilitate decision-making in prescribing.
Artificial intelligence (AI) needn’t be seen as the downfall of human-interaction medical care. It is a means of aiding the physician or healthcare worker in doing their job faster, updating EHR, and providing more options for care and solutions to be considered.
Meskó predicts that AI will take its rightful place in healthcare with computers helping physicians in clinical decision-making. He saw it in the realm of the stethoscope’s acceptance decades ago.
Diagnosis and prescriptions are two areas where AI can be of considerable use. However, AI does have its niches. Prescribing may be in the area of AI known as Narrow AI where a specific task is needed.
CME Prescribing
For example, one commentator concludes that: “Continuing medical education has become so heavily dependent on support from pharmaceutical and medical device companies that the medical profession may have lost control over its own continuing medical education.”
CME prescribing presentations provide niche information, but it is the prescriber's responsibility to know when there’s an inappropriate combination, especially in a geriatric population. AI may help here.
As the population ages and the numbers of patients in the over 65 category increases, it is an especially relevant, rapidly emerging issue.
Starting in 2030, when all boomers will be older than 65, older Americans will make up 21 percent of the population, up from 15 percent today.
By 2060, nearly one in four Americans will be 65 years and older, the number of 85-plus will triple, and the country will add a half million centenarians.
With 500K patients over 100 years old, a population of this size will present surprisingly unexpected difficulties in health maintenance. How many medications will they need? How much and which medications will their bodies be capable of adequately metabolizing?
Polypharmacy may be problematic for a number of reasons. For example, it may increase the risk of using potentially inappropriate medications (PIMs), which have been associated with negative effects on long-term physical and cognitive functioning.
In more instances than not, elderly patients may develop iatrogenic illnesses, and, as described in a recent article, it begins benignly.
The glide path to overuse can be gradual: A patient taking a drug to lower blood pressure develops swollen ankles, so a doctor prescribes a diuretic. The diuretic causes a potassium deficiency, resulting in a medicine to treat low potassium. But that triggers nausea, which is treated with another drug, which causes confusion, which in turn is treated with more medication.
Would a computer in a prescriber’s office enable prompt attention to this internal avalanche of medications and their attendant effects? Able to quickly track and assess patient records, AI programs can outpace providers to determine appropriate medications or combinations. It is here that it is most valuable.
Pocket Prescribers to the Rescue?
Not too long ago, RAM concerns in cell phones was solely for those who save photos, music, or texts. Today, prescribing programs point to either more memory or add-on to phones to store data for patients and prescribers.
But memory in a cell phone isn’t the only consideration. The major pros and cons assist prescribers in signaling error potential as well as contraindications for medications. There are also additional concerns requiring review when prescribing.
The outcome of one study was revealing and concerning to the ISMP.
The study reviewed more than 26,000 free text notes included with ambulatory care prescriptions and found that more than 66% contained inappropriate content for the prescription, 29% contained appropriate content, and 5% contained information considered unnecessary. The study appeared in JAMA Internal Medicine.
The increasing complexity as well as genetic considerations which are required in prescribing software point to an ever-increasing need for its use. Failing to contend with AI over human fallibility can be fatal.
But what about the potential errors in these software programs? Ongoing research in medical algorithms is one answer to these issues. Even computers can make coding errors when the data is problematic.