I’ve long believed that, just as no two psychiatric patients are identical, there is– and never will be– a “one size fits all” approach to psychiatric care. However, much work has been done in the last several years to develop “algorithms” to guide treatment and standardize care. At the same time, the adoption of electronic health record (EHR) systems– which are emphasized in the new U.S. health care legislation– has introduced the possibility that computerized decision-support systems will help guide practitioners to make the right choices for their patients. It is my opinion that such approaches will not improve psychiatric care, and, in fact, will interfere with the human aspect that is the essence of good psychiatric practice.
“Clinical decision support,” or CDS, is the idea that an algorithm can help a provider to give the right kind of care. For a busy doctor, it makes sense that getting a quick reminder to prescribe aspirin to patients with coronary artery disease, or to give diet and exercise recommendations to patients at risk for obesity or diabetes, helps to ensure good care. Several years ago, I actually helped to develop a CDS system designed to remind primary care doctors to avoid opiate painkillers (or use them with caution) in patients who had a history of substance abuse or other relative contraindications to narcotics. At the time, I thought this was a great idea. Why not harness the ability of a computer to gather all the data on a given patient– something that even the best doctor cannot do with absolute accuracy– and suggest the most advisable plan of action?
Now that I spend most of my time actually practicing medicine, and using two different EHR systems, I’m having second thoughts. While I appreciate the ability to enter patient data (and my notes) into a system that is instantly accessible by any provider in my office at any time, and write prescriptions with a few clicks of my mouse, I’ve begun to resent the ways in which EHRs tell me how to practice, particularly when (a) they give recommendations that I would employ anyway (thereby wasting my time), or (b) they give recommendations that deviate from what I believe is right for the patient.
Obviously, the latter complaint is particualrly relevant in psychiatry, where each patient presents a different background of symptoms, stressors, preferences, and personal history. When anyone asks me “who is your ideal patient for drug X?” or “what is your first choice of drug for depression?” I find it hard to give an answer. Treatment choices come down to a feeling, a gestalt, incorporating both observable data and intuition; it’s hard to describe and impossible to quantify.
One example of a psychiatric CDS is based on the Texas Medication Algorithm Project (TMAP). The TMAP was developed to help providers determine what medications to use in the treatment of mood disorders; the first version of TMAP for depression was designed in 1999 and implemented in a computerized CDS in 2004. A pilot study involving four primary care providers, published in 2009, showed that depression outcomes were slightly better (i.e., scores in the HAM-D were lower) in the group using the CDS. (This may have been due to the setting; in a busy primary care clinic, any guidance to address depression symptoms may improve outcomes relative to no guidance at all.) However, a follow-up study by the same group found that it was much harder to implement the CDS on a more widespread scale in mental health clinics, due to technical problems, poor IT support, billing & coding problems, formulary issues, recommendations that providers disagreed with, lack of time, and impact on workflow.
That may have been for the better. A new study in this month’s Archives of Internal Medicine by Romano and Stafford shows that CDSs may just be a waste of time and money. They evaluated over 330 million ambulatory care patient visits using EHRs over 2005-2007, 57% of which involved at least one CDS, and found that, on 20 quality-of-care indicators, using a CDS contributed to improvements in treatment (i.e., treatment concordant with established guidelines) on only one measure. (Two measures involved psychiatric conditions– one was for the treamtent of depression, and the other was to remind providers not to use benzodiazepines alone for depression treatment. Neither of these measures showed improvement when a CDS was used, relative to no CDS.)
So despite all the resources devoted to electronic medical records and clinical decision support systems to improve care, the evidence seems to indicate that they don’t. Either doctors ignore CDSs and provide “practice as usual” anyway, or the CDSs give recommendations that doctors already follow.
This may be good news for psychiatry, where treatment guidelines (thankfully) offer a great deal of latitude, but CDSs, by their very nature, may restrict our options. In the future, then, when we believe that the patient sitting in front of us is a good candidate for Effexor, or Seroquel, or interpersonal therapy with no meds at all, we may no longer need to explain to a computer program why we’re ignoring its recommendation to try Prozac or Haldol first.
In my opinion, anything that preserves the integrity of the physician-patient interaction– and prevents the practice of medicine from turning into a checklist-and-formula-based recipe– preserves the identity of the patient, and improves the quality of care.
Addendum: See also a related post today on 1boringoldman.com.