In a World With AI, How Will Our Care of Patients Be Measured?
Barbara J. Howard, MD
Medscape February 12, 2026
Have you ever worried that listening to a parent’s long story about being up all night with two kids vomiting and their spouse away on a business trip will be measured as poor quality care? It depends on how quality is measured in our evolving health system.
Electronic health records (EHRs) were inserted into our offices with payment incentives (carrots and later sticks) to enable data acquisition and thereby improve healthcare. Much of what followed has been a tyranny of boxes to check and templates purported to measure the care provided and pressure to conduct more tests in part to demonstrate thoroughness and reduce liability.
The next stage of the development of measurement has been the advent of artificial intelligence (AI), which can collect all the past and present data in the EHR for each patient, extracting meaning from our notes in the history section — including information entered by a human or AI scribe listening to the visit conversation. These data are now beginning to be used to determine diagnoses and their level of severity, measure our completion of related guideline-based care, document appropriate treatment, and create a care plan.
The AI program can suggest diagnoses and insert billing codes to optimize payment. It can create preauthorization letters and send patients educational materials. Much of this AI assistance may be welcome as documentation has come to take up 65% of our time as clinicians. It is easier to review what AI has written than to generate it ourselves. And we certainly want to be sure we are providing quality care. We may sometimes use AI suggestions and decision support. What could possibly go wrong?
The parts of the healthcare visit that can be readily measured are likely to be the basis of conclusions about the quality of our care. But as we are eliciting the information that national guidelines require for diagnosis, determining severity, and recommending treatment, are we being empathic listeners during the brief visits?
Asking families to complete pre-visit questionnaires that request the detailed information required to fulfill guideline criteria for their condition can make some of this easier. When we thank families and give examples of how this work on their part contributes to accurate and comprehensive care, we are truer partners in their care.
This also makes time for shared decision-making that is respectful of their autonomy and enables individualized treatment approaches based on guidelines. The documentation, then analyzed by AI, can more accurately determine whether criteria were met and appropriate treatment advised.
Perfect, right? But our clinical observations, relationships, and interactions with the family members, our knowledge of the community context, and what we have learned about the patient from previous visits or phone calls may only be covered in guidelines by such instructions as “conduct psychosocial assessment.” We know, and data show, that this information is frequently as important or more important to diagnosis, treatment, patient adherence, family satisfaction, attendance at follow-up visits, and functional outcomes as numerical data.
So, how can we be assured that this clinician-derived information is also measured when quality is being assessed and guidelines are being updated?
Important research has been done to examine the aspects of clinician-family interaction that benefit care outcomes and should be included in quality measures. Because 25%-50% of health supervision visits note the child’s behavior or development as the main concern, it is important to determine how to measure factors that influence outcomes for these issues.
Measuring Behavioral Diagnoses and Treatment
Dr Larry Wissow and colleagues studied “common factors” that have value in dealing with behavioral health challenges in primary care. Although the main goals in studying these factors have been to empower primary care clinicians with these basic skills and to optimize patient outcomes, another timely benefit is that they are discrete factors that have broad applications — enabling us to measure the of quality of these processes.
The common factors approach is designed to initiate care immediately without a specific diagnosis or specialty referral for a broad range of conditions and to be implemented in primary care. The approach has shown positive effects on outcomes including maternal mental health and child functional status.
The key domains in the common factors paradigm are clinician-patient relationships (with both the parent and the child) and clinician skills that influence patient behavior change. A positive clinician-patient relationship is promoted mainly by “demonstrating empathy, warmth, and positive regard” rather than the clinician being affectively neutral. The skills shown to influence behavior change include “clearly explaining the condition and its treatment, keeping the discussion focused on immediate and practical concerns, and keeping the treatment session organized.”
Beginning treatment without a definite diagnosis, core to the common factors approach, is consistent with primary healthcare as well as behavioral healthcare. In every case, the common factors approach includes interacting with the family to elicit information on how they see the problems, expressing caring as well as optimism about the concerns, and developing a working relationship with the family and child that promotes family problem-solving, use of resources, and behavior change.
In many cases, these defined interactions result in reduced symptoms without other interventions and a willingness to participate in the care process. The training for primary care clinicians that showed a positive effect was as little as 4 hours over several weeks.
There are other measurable “practice elements” that also fall within a small knowledge set that primary care clinicians could measurably implement to address more specific behavioral health conditions. Of course, we need to recognize emergent issues for referral (eg, suicidal ideation, psychosis). But there are also specific skills we can apply for the most common general categories of mental health disorders in children: attention-deficit/ hyperactivity disorder (ADHD), anxiety/avoidance, depression/withdrawal, and disruptive behaviors.
For example, in addition to medication management, behavioral techniques with evidence for success with ADHD include parent education (58%) and coaching on parental use of praise (83%), rewards (92%), monitoring (83%), time out (83%), response cost (58%), and commands and limit setting (58%).
This same set of skills has evidence for benefit in managing disruptive or willful misbehavior and depression. Motivational interviewing, another evidence-based practice, can also be learned, implemented, documented, and billed for in primary care (eg, for substance use).
There are other published methods for measuring doctor-patient communication to consider. The STANDARD project has been proposed to collect uniform data for neurodevelopmental and behavioral conditions including “diagnostic criteria, psychosocial and behavioral treatment, and interventions and response” as the current lack of standardization limits “large-scale research to improve the evidence base for diagnosis and treatment.”
The Roter interaction analysis system is the most widely used method, but it requires human observation and coding. Tools completed by patients for assessing clinician-patient interaction are fraught with complex individual issues but some have been validated as relating to outcomes, including the Working Alliance Inventory, patients’ Perceived Involvement in Care Scale, and the Trust in Physician Scale.
Now that clinicians are using AI scribes to transcribe voice to text, it may be that measurement of our interactions could be derived from analysis of these notes. AMIE (Articulate Medical Intelligence Explorer), a large language model-based AI system, was taught to produce text conversations that “not only had greater diagnostic accuracy than the physicians but also patient-actors and specialist raters both evaluated AMIE’s performance to be higher than that of the [primary care physicians] on metrics related to empathy and communication skills.”
The AMIE model worries me about our future, but it does not mean that AI can provide better care than we can in the brief in-person encounters we have with patients, at least not yet. But we need to consider what sort of measurement may be needed to maintain respect for our clinical skills.
Medscape Pediatrics © 2026 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: In a World With AI, How Will Our Care of Patients Be Measured? - Medscape - February 12, 2026.