Outcome tracking may support retention when clinicians receive the results, discuss them with patients, and respond to emerging problems. Measurement alone does not guarantee that someone stays in care. Define dropout, planned completion, transfer, and missing follow-up before testing whether a feedback workflow changes retention in your setting.
What the evidence supports
The strongest relevant evidence concerns progress feedback, not score collection by itself. A 2021 multilevel meta-analysis included 58 studies, 110 effect sizes, and 21,699 patients. It found a small favorable effect on symptoms and a small favorable effect on dropout.
The dropout estimate was an odds ratio of 1.19, with a 95% confidence interval from 1.03 to 1.38. That result does not mean every program improves retention by 19%. Odds ratios are not percentage-point changes, and the studies varied in measures, feedback, frequency, treatment intensity, and setting.
The intervention included feedback. A questionnaire that is delivered, scored, and filed is not the same thing. The APA's 2025 measurement-based care guidelines frame MBC as clinical competencies that include collecting, sharing, and using data during care.
Treat retention as an outcome to test, not a promised return.
Feedback is the intervention.
Define retention before looking at the result
"Stayed in treatment" is too vague for analysis. Choose an endpoint that fits the service:
- Attended the next planned visit
- Remained active through a defined review date
- Completed an agreed episode or program
- Ended care by mutual plan
- Transferred to another service
- Stopped attending without a planned ending
Do not combine planned completion with disengagement. Do not count transfer as failure unless the contract says so. Keep patient-directed endings visible rather than forcing them into success or dropout.
Labels shape the result.
Define the denominator at the same time. It might include all new episodes, everyone with an initial assessment, or everyone expected to receive follow-up. Each choice answers a different question.
Separate the workflow stages
A retention analysis should distinguish at least five stages:
- Assessment became due.
- Delivery was attempted through the intended channel.
- The patient completed a usable assessment.
- The responsible clinician reviewed it on time.
- The result was discussed or otherwise used in care.
If retention changes, these fields help identify what actually changed. A practice cannot attribute the result to feedback when few clinicians reviewed the scores. It cannot call missing assessments successful monitoring.
The guide to automated assessment scheduling explains why delivery, completion, review, and clinical action need separate states.
Use scores to open a conversation
Repeated PHQ-9 or GAD-7 results can give a clinician and patient a shared reference point. The useful question is not whether the graph proves treatment worked. It is whether the pattern helps them discuss symptoms, function, goals, barriers, and fit.
A neutral conversation might sound like this:
> Your answers changed since the last visit. Does that match your experience? What feels different, and what does the score miss?
The patient may describe practical barriers, concern about the care plan, an adverse effect, a change in symptoms, or no meaningful change at all. The score does not identify the cause.
Context still decides.
Do not use improvement to pressure someone to continue. Do not use a low score to imply that care is no longer needed. Retention is not automatically good when care lacks benefit, and an appropriate planned ending is not failure.
Route clinical concerns separately
The retention workflow must not absorb urgent clinical review. Item-level safety signals follow the practice's safety protocol regardless of the total, trend, appointment status, or retention metric.
A missed assessment is also not a clinical conclusion. Record noncompletion and apply the practice's outreach rules. Do not infer stability, improvement, or risk from silence.
Silence is missing data.
Build a reproducible retention report
Start with a cohort flow:
| Count | Definition |
|---|---|
| Eligible episodes | Met the prespecified service and date rules |
| Feedback eligible | Had a usable result that could reach the clinician |
| Feedback reviewed | Clinician review met the defined timing rule |
| Planned endings | Completed, transferred, or ended by agreement under the protocol |
| Unplanned endings | Met the prespecified disengagement definition |
| Unknown | Outcome could not be classified from available data |
Compare groups only when the design supports it. Patients who complete assessments may differ from those who do not. Clinicians who use feedback may differ from those who do not. A before-and-after practice report can identify a local pattern, but it cannot remove those sources of bias by itself.
Show assessment coverage and missingness beside retention. The broader guide to measuring treatment effectiveness explains cohort, episode, and denominator controls that apply here too.
Report both.
Test the process before assigning financial value
Measure the existing retention process first. Then implement the full feedback workflow in a defined service line. Track training, assessment completion, clinician review, patient discussion, planned endings, unplanned endings, and unknown outcomes.
Prespecify the evaluation period and decision rule. Look for operational problems as well as the retention result. A workflow that adds burden, delays urgent review, or leaves many results unseen has not succeeded even if the headline rate changes.
Do not convert retained visits directly into revenue without checking capacity, patient benefit, and contract terms. The business-case guide shows how to keep measured operational value separate from assumptions.
Outcome tracking can support a better clinical conversation and may modestly support retention when feedback is actually used. The defensible claim is narrower than the original promise. Measure the full workflow, distinguish appropriate endings from disengagement, and let the local data show whether the process helps.