ADDRESS PREVENTIVE CARE GAPS AT ACUTE VISITS

To improve our clinic’s performance on preventive care metrics, we implemented a protocol to discuss preventive care at acute care visits. To facilitate this, our EHR displays a “Best Practice Alert” when preventive care gaps exist in the patient’s record. This prompts rooming staff to use the following script: “In addition to (clinician) seeing you for (acute concern), it appears you have some preventive care due, specifically (preventive services). Would you like to schedule a separate appointment to discuss the preventive care, have (clinician) discuss it at this visit, or discuss this at another time?”

If patients opt to discuss it that day, rooming staff click on the EHR alert. With one click, care gaps become pending orders and associated diagnoses are added to the visit. Clinicians then add the preventive care discussion to the visit agenda. After the first six months, this system produced significant gains in colon and breast cancer screening, winning over even our clinic’s skeptics.

EXERCISE CAUTION WHEN USING AI TOOLS LIKE CHATGPT

Chatbots such as ChatGPT and Google BARD can play a role in medical communication, but exercise caution with this technology because of the risk of misleading or incorrect information. We conducted an experiment last year, feeding ChatGPT a series of questions about common health conditions and then assessing its accuracy and coherency. For example, we asked what percentage of people experience nausea with a concussion?

ChatGPT’s response: “The prevalence of nausea in people who have a concussion varies depending on the study, but it is a common symptom. According to one study, up to 40% of people with a concussion experience nausea...”

We asked for more information about the 40% figure.

ChatGPT’s response: “… the study I was referring to is ‘Acute Symptoms and Time to Recovery Affer Sport-Related Concussion in Children and Adolescents’ by William P. Meehan III, Rebekah C. Mannix, and R. Dawn Comstock. It was published in the journal JAMA Pediatrics in 2013.”

We searched for this article but were alarmed to discover it did not exist. ChatGPT fabricated an article citation and summary. There are published cases of this phenomenon — AI “hallucinations.”¹ Asking the program the same concussion question four months later yielded a different answer; it no longer cited the 40% figure from the non-existent study. The software constantly learns and improves, but clinicians should understand its shortcomings and caution their patients.

OVER-RECRUIT PATIENTS FOR SHARED MEDICAL APPOINTMENTS

Our study of shared medical appointments for patients with type 2 diabetes in 22 practices found that one key for making them financially feasible was to recruit more patients than needed to hedge against no-shows. Most practices found that the ideal group size was about 15 patients, but to hit that number consistently they had to recruit up to 25.

Successful groups began recruiting at least eight weeks in advance. A nurse or staff coordinator usually led recruiting efforts, but the most effective strategy was a clinician invitation. This was true even if a patient’s clinician was not conducting the shared visits. To save clinicians time, other team members can reach out to patients but mention that the clinician thought of them for participation. Practice staff can use the EHR to identify candidates — for example, patients with A1C>9, patients who have not had eye or foot exams in 12 months, or newly diagnosed patients.

Practices can also recruit passively with flyers in waiting rooms, automated messages in patient portals, and advertisements in traditional or social media. More tips are available in the file available for download here.