If one challenge defines diabetes care, it is treatment discontinuation.
What makes it difficult is that it is hard to notice.
Clinics see many patients daily. Physicians and staff are fully occupied with those in front of them. In that state, patients who did not come are invisible. Nobody thinks "Mr. A didn't come today."
The longer a condition runs without symptoms, the more motivation patients have to lapse. "I feel fine and I'm busy—I'll skip this one." From there, six months pass, then a year.
Discontinuation is not an accident; it progresses quietly. That is precisely why it must be detected by mechanism rather than attention.
Disclaimer: This article provides general information. Clinical policy follows each clinic's judgment.
What Counts as Discontinuation
Before building the mechanism, the detection definition must be set. Several entry points exist.
1. Detection by visit interval
The most direct approach: no visit after a set period since the last, or no next booking (or a no-show left unresolved).
But diabetes intervals differ by patient—monthly, bimonthly, quarterly. A uniform "three months means discontinued" rule misses patients whose normal interval is three months.
Deviation from each patient's usual interval yields better precision.
2. Detection by prescription expiry
Practically more powerful.
- How many days the previous prescription covered
- No visit after that period elapsed
An expired prescription means medication is likely not being taken—a more direct indicator than visit interval.
3. Combining lab deterioration with attendance
- HbA1c worse than last time, with no next booking
- Sustained deviation from targets alongside a widening interval
The idea is prioritizing those who will deteriorate if left alone. Given that resources to chase everyone equally do not exist, this provides prioritization.
4. Incomplete assessments
Diabetes care includes periodic assessments: ophthalmology referral and attendance confirmation, renal and urine testing, foot examination, nutritional guidance.
Items scheduled but not performed are also a follow-up entry point.
What the System Must Do
Can it extract the cohort?
Everything starts here.
- Can patients be listed by the above conditions?
- Can conditions be configured and changed in-house?
- Does extraction run automatically on a schedule?
- Can response status be recorded against results?
The last is easily overlooked. Extraction alone is not work. Only when "contacted / unreachable / attended" can be recorded does the next action follow.
Can it handle lab time series?
- Are HbA1c, glucose, weight, and blood pressure held as structured data?
- Can they be graphed over time?
- Can deviation from targets be evaluated?
Values buried in free text cannot serve as extraction criteria at all. The same structural issue appears elsewhere (see EMRs for Neurology Clinics, EMRs for Urology Clinics).
Are there contact channels?
- Are reminders (SMS, email, app notification) supported?
- Are send results recorded?
- Can phone contact be logged?
Glucose data import
For patients using self-monitoring or continuous monitoring, data sits on the device.
- Can device and app data be imported into the chart?
- Is the between-visit picture visible?
Seeing the interval between visits can reveal early signs. See EMRs for Diabetes and Endocrinology Clinics.
Checklist
| Area | Items |
|---|---|
| Extraction | By interval / prescription expiry / lab deterioration / incomplete items |
| Configuration | In-house editing / per-patient interval / scheduled automation |
| Recording | Response status / contact history / return tracking |
| Lab values | Structured storage / time-series graphs / target deviation |
| Contact | Reminder sending / send results / phone logs |
| Devices | Glucose import / between-visit visibility |
Designing the Outreach
Once extraction works, the next question is who contacts patients, and when.
Introduced without deciding this, the list appears and nobody acts. Deciding these in advance makes it move:
- Assign responsibility (nurse or administrative staff)
- Set frequency (weekly, monthly)
- Set a ceiling (how many cases per round)
- Prepare message templates (do not compose each time)
- Decide how to handle repeated non-response
The third matters most. Trying to contact everyone is unsustainable. A fixed number, highest priority first, keeps the operation running.
AI-Native as an Option
- Prioritization: ranking extracted patients by risk
- Message drafting: composing outreach tailored to each situation
- Inquiry handling: automatically receiving rescheduling contact
- Course summarization: grasping a patient's situation briefly before calling
- Trend analysis: understanding which patients tend to lapse
The last leads to designing preventive intervention. If "dropout concentrates between the first and third visit" becomes visible, engagement during that window can change.
The premise that AI output is not sent to patients unreviewed remains necessary. See Where to Draw the Line on Delegating Work to AI.
Conclusion
- Discontinuation is the central challenge, and the core problem is that patients who did not come are invisible
- Four detection entry points: interval, prescription expiry, lab deterioration, incomplete items
- Intervals differ by patient, so deviation from an individual's usual interval beats a uniform threshold
- Prescription-expiry detection directly indicates that medication has run out
- Extraction alone is not work; recording response status must come with it
- Values buried in free text cannot even serve as criteria
- Operations stall unless responsibility, frequency, ceiling, and templates are decided first
For details on AI Karte or to request a demo, please contact us.