When the Healthiest Patients Quietly Leave: A Two-Cohort Study of Screening Return at a Single Breast Imaging Practice

Mammologix Breast Imaging Outcomes Intelligence · Study Report · First in the Tier 1 Retention Series

Data cutoff: May 30, 2026. Single de-identified facility. Retrospective analysis; not peer-reviewed.

---

Summary

A breast imaging practice told 3,686 women, after a normal mammogram in 2022, the same thing it tells nearly every healthy patient: come back in a year. This study followed every one of those women forward through May 2026 to answer a question the field measures poorly — namely whether they returned, and on time. The same analysis was then repeated on an independent 2023 group of 1,940 women. Two findings stand out. Among the women who came back, the annual rhythm is genuinely strong. And a large, quiet group did not come back at all, most of them disappearing from the practice entirely rather than moving into other care. The pattern held its shape across both independent years, which is what separates a finding from a one-year accident. This is the first study in a series: an introduction to the method and the early results, with deeper cohorts and an outcomes analysis to follow.

---

1. Why This Study Exists

Screening mammography earns its value through repetition. A single normal mammogram, read perfectly, does little on its own; the benefit accrues only when a woman returns year after year so that change can be caught early. The evidence for that is not subtle. A prospective study of roughly 549,000 women found that those who attended their two most recent scheduled screens before a diagnosis had about half the breast cancer mortality of non-participants, and that missing even one of those two screens measurably raised the risk.[2] Population data point the same direction on stage: in one large analysis, women who never attended screening were nearly six times more likely to present at an advanced stage than regular attenders, and even irregular attenders carried higher risk than those who came on schedule.[3] Regular return is not an administrative nicety. It is most of where the lifesaving happens.

Set against that, the field has a measurement blind spot. Breast imaging quality assessment is mature in one direction: the Mammography Medical Outcomes Audit (MMOA), required under the Mammography Quality Standards Act (MQSA), the federal law administered by the U.S. Food and Drug Administration, tracks positive exams through to tissue diagnosis and reports cancer detection rate, recall rate, and positive predictive value against national norms. The MMOA grades the accuracy of the reads. What it does not measure is whether the normal, asymptomatic patient comes back. National surveys tell us roughly 80 percent of women aged 50 to 74 report a mammogram within the past two years, with lower figures among younger and lower-income women,[1] but that is a cross-sectional snapshot, not a record of whether a given practice's own patients return on a real annual cadence over time. Most practices cannot answer that question for themselves. This study was built to answer it.

The women in this study share one starting point: each came in for a routine, asymptomatic screening mammogram, received a normal result (negative or benign), and was advised to return to annual screening. This study calls that group Tier 1. They are the practice's healthiest and most routine patients — the ones for whom an annual return is both expected and appropriate. A woman can move in and out of Tier 1 over time. If she returns the next year and is again normal, she stays in Tier 1. If she returns but is recalled for additional imaging or arrives with symptoms, she is recorded as seen, but not at Tier 1 that year, and she can re-enter later. That distinction turns out to matter a great deal.

2. What We Did

Two independent entry cohorts were built from a single de-identified facility extract of roughly 24,000 imaging events spanning January 2022 through May 30, 2026. The 2022 cohort is every woman whose first qualifying Tier 1 screen fell in 2022 (3,686 women). The 2023 cohort is every woman whose first qualifying Tier 1 screen fell in 2023, with the 2022 entrants deliberately excluded so the groups do not overlap (1,940 women). Each cohort was then followed forward to the cutoff. Before any new analysis, the Tier 1 definition was applied to the raw events and shown to reproduce the practice's own existing Tier 1 extract exactly — down to the same six patients with two same-day qualifying screens — which confirmed the pipeline was faithful to the practice's logic before anything was built on it.

The method borrows from clinical research rather than from routine reporting, in five ways worth naming:

Return is modeled as time-to-event. Using the Kaplan-Meier (survival) method, each woman stays in the denominator only for the time she has actually been eligible to return. A woman screened in late 2025 is not a non-returner in early 2026; she is simply not yet due. This censoring correction is the honest way to measure return behavior, and it matters more for the younger 2023 cohort, where more women are legitimately not yet due.

Compliance is a named spectrum, not a pass-or-fail line. Return intervals are sorted into bands by how long the woman took: strict annual (10.5 to 13.5 months), semi-strict annual (13.5 to 16.5 months), loose annual (16.5 to 21 months), biennial (21 to 33 months), and extended lapse (beyond 33 months).

The bands are the return curve, sliced. The bands are not a separate scoring scheme. They are the censoring-adjusted curve cut into labeled segments, which is what keeps the percentages honest as newer cohorts mature.

The genuinely lost are separated from the actively managed. A woman who stops appearing at Tier 1 has either been escalated into other care (recalled, worked up, symptomatic) or has vanished entirely. Most retention measures blur these together. This one separates them.

Cross-cohort claims use matched horizons. Because the 2023 cohort has had a year less to return, every comparison is matched to the same number of months since entry — the only fair way to ask whether retention is holding or slipping.

3. What We Found

Among the women who returned, the cadence is strong. In the 2022 cohort, 71.7% returned for at least one Tier 1 screen, with a median first return near 18 months. Of all return intervals, 63.8% were strict-annual and better than three in four fell within roughly fifteen months. The practice is not fighting chronic lateness among the patients who stay.

The problem is not lateness. It is disappearance. In the 2022 cohort, 28.3% never returned for a Tier 1 screen across more than four years. More telling, 948 women — about one in four of the entire cohort — were never seen at the practice again in any capacity. Only 96 of the non-returners had moved into other care. The dominant reason a woman leaves Tier 1 is that she leaves the practice, not that the system caught something and redirected her.

The pattern repeats across two independent years. The 2023 cohort shows the same shape: 66.0% returned at least once so far — a figure that will keep rising as the cohort matures — and 591 women (30.5%) have not been seen again at all. Matched to the same clock, the two cohorts track closely. By 33 months since entry, 67.2% of the 2022 cohort and 64.8% of the 2023 cohort had returned at least once. The 2023 group is slightly slower and slightly less strictly on-time, but the fundamental structure is the same. A strong on-time culture for those who stay, paired with a persistent silent-attrition problem, is now a documented feature of this practice across two years rather than a fluke of one.

Measure (matched where follow-up windows differ)	2022 Cohort	2023 Cohort
Women entering Tier 1	3,686	1,940
Returned at least once (within window)	71.7%	66.0% (still rising)
Returned at least once by 33 months	67.2%	64.8%
Never seen again at the practice	948 (25.7%)	591 (30.5%)
Strict-annual share of returns	63.8%	55.8%
Past recommended return date, no screen on record	2,103	1,094
Of those, more than 15 months overdue	1,368	680

The overdue group is sizable, and it is the part a practice can act on. As of the cutoff, 2,103 women in the 2022 cohort and 1,094 in the 2023 cohort were past their recommended return date with no later screen on record. This analysis runs on de-identified data and identifies no one; these figures describe how many patients fall into the overdue category, not who they are. The point is not these particular women. It is what their numbers represent: silent attrition at this scale is measurable, it is sizable, and it is the kind of pattern any breast imaging practice can look for in its own population. When a practice applies the same method to its own records — where patients are of course identified and where the practice is permitted to use that information for routine care and outreach — the output becomes a prioritized recall list it can act on.

The clinical stakes are real, and stated carefully. Within the 2022 cohort, 42 women received a malignant diagnosis during the window; within the 2023 cohort, 19 did. These figures are presented as context and stakes, not as a proven consequence of any specific lapse. This study does not establish that a missed return caused a delayed diagnosis, and it does not attribute these diagnoses to gaps. What the broader literature does establish, separately, is that the women most exposed to late-stage disease are disproportionately the ones who attend irregularly or stop attending.[2][3] Connecting this practice's own lapse patterns to stage at detection is a dedicated outcomes study — and it is a planned next step, not a claim made here.

4. Why This Matters, and What Can Be Done With It

This kind of retention intelligence does not replace the MMOA; it complements it. The MMOA grades the accuracy of the reads. Retention intelligence measures whether the screening relationship persists over time, patient by patient — the dimension the standard audit was never designed to capture.

The encouraging part is that the failure mode here is addressable, and the tool to address it is evidence-based. The Community Preventive Services Task Force recommends provider reminder and recall systems on the strength of consistent evidence that they increase breast cancer screening, with reminders flagging patients who are due and recall targeting those who are overdue.[4] A prioritized list of overdue patients is exactly the input such a program needs. The recently slipped are the most recoverable with a simple prompt; the deepest-lapse group carries the highest clinical risk and warrants direct contact. The recovered screening volume that follows is a modeled opportunity that depends on each facility's own reimbursement rates — not a realized figure — but the patient-safety case stands on its own.

A word on scope, because credibility depends on it. This is a single de-identified facility, analyzed retrospectively, and it has not been peer-reviewed. It is a strong proof of concept and a repeatable method, not a published clinical trial. Survival analysis of screening adherence is itself well established in academic literature. What is unusual here is the delivery: a research-grade method applied to routine operations, run identically on independent cohorts, and ending in an action rather than a slide. The method is designed to run at the level of an ordinary breast imaging practice, on the data that practice already holds — not only in an academic center. That is what makes it adoptable rather than merely admirable.

5. What Comes Next

This is the first study in the series. It establishes the method and the first two years of evidence. The series will widen in three directions. Additional entry cohorts, beginning with 2024 and 2025, will sharpen the trend line and show whether the modest softening from 2022 to 2023 is noise or a signal worth managing. The higher follow-up tiers — where patients are recalled, worked up, or sent to surveillance and biopsy — will show how the practice manages its more complex cases. And the outcome-linkage study will connect lapse patterns to stage at detection, the analysis that would turn an operations finding into a patient-safety case. Each additional year and tier runs on the same engine at low marginal cost, which is what makes a longitudinal, cohort-over-cohort quality picture practical rather than aspirational.

The single normal mammogram is the beginning of a screening relationship, not the end of one. This study is a definitive step forward in measuring whether that relationship holds — for real patients, at a real practice — and in giving a practice a way to find the patients who slip away while there is still time to reach them.

---

Full Study Reports

The complete technical reports — including all Kaplan-Meier curves, compliance-band charts, attrition breakdowns, and full methodology details — are available as PDFs below.

2022 Entry Cohort (10 pages) View report → · Download PDF

2023 Entry Cohort — Companion Report (11 pages) View report → · Download PDF

---

References

1. Centers for Disease Control and Prevention. Use of Cancer Screening Tests, United States, 2023 (National Health Interview Survey). Prev Chronic Dis. 2025. (≈80% of women aged 50–74 up to date with mammography within 2 years; 62.1% of women aged 40–49.)
2. Duffy SW, Tabár L, Yen AMF, et al. Beneficial effect of consecutive screening mammography examinations on mortality from breast cancer: a prospective study. Radiology. 2021;299(3):541–547. doi:10.1148/radiol.2021203935 (≈49% lower breast cancer mortality among serial participants vs non-participants; added benefit from attending both of the two most recent screens vs one.)
3. Population-based study of screening regularity and stage at diagnosis, Flanders, 2022 (women aged 50–69, 2001–2018). Advanced-stage risk vs regular attenders: irregular OR 1.17, attended-only-once OR 2.18, never-attenders OR 5.95. [Full citation to be completed before external distribution.]
4. Community Preventive Services Task Force. Cancer Screening: Provider Reminder and Recall Systems — Breast Cancer. The Community Guide (CDC). (Recommended based on strong evidence of effectiveness in increasing breast cancer screening by mammography.)

Methodology Notes

Longitudinal entry-cohort follow-up with Kaplan-Meier (censoring-adjusted) return modeling, using two curves (time to first return, and cadence between consecutive screens, with patient-level clustering on the latter), a named-band compliance spectrum read off the adjusted curve, explicit separation of lapsed from escalated patients, and matched-horizon cross-cohort comparison. A woman is counted past her recommended return date when her most recent Tier 1 screen plus its recommended follow-up date precedes the May 30, 2026 cutoff with no later qualifying screen. The 2023 cohort's shorter window means its never-returned and extended-lapse figures are under-observed relative to a mature cohort and should be read as lower bounds.

Disclosures

Mammologix is a commercial offering of I/O Trak, Inc. The author and publisher may benefit commercially from adoption of the method or related software workflows described here. This report was prepared internally and received no external funding.

All data in this study are de-identified, and the study identifies no individual patient. Any future version incorporating client-derived figures will use aggregated, de-identified, contract-permitted data that has undergone privacy review, including applicable HIPAA requirements, prior to external distribution.

---

BI-RADS® is a registered trademark of the American College of Radiology, used here for educational purposes only. BI-RADS® assessment categories are referenced implicitly throughout this study; "negative or benign" Tier 1 criteria correspond to BI-RADS® categories 1 and 2, and "recalled for additional imaging" corresponds to BI-RADS® category 0.