Methodology & evidence
Ampulert combines published associations between glycemia and lower-extremity amputation with transparent, versioned modeling assumptions. The result is an educational five-year population estimate—not a validated prediction of what will happen to an individual patient.
Research-informed MVP model. External validation and recalibration are still required.
What the number means
The displayed percentage estimates the probability of any nontraumatic lower-extremity amputation over five years in a general diabetes population without a current active foot problem.
It is not a diagnosis, medical advice, a statistical confidence interval, or a validated individualized clinical-prediction score.
Ampulert should not be used for a current ulcer, infection, gangrene, ischemic emergency, or newly cold or pale foot. Those conditions require prompt clinical assessment.
Calculation
When A1C results are available, Ampulert uses up to the six most recent values. Newer results receive more influence through exponential recency weighting with a 180-day half-life.
weight = 0.5^(age in days ÷ 180)If no A1C is available, up to 14 glucose readings are weighted with a 30-day half-life and converted using the published ADAG relationship.
A1C proxy = (average glucose mg/dL + 46.7) ÷ 28.7The MVP begins with an Ampulert-selected five-year anchor of 1.7% at an effective A1C of 7.0%. That anchor is an extrapolated population assumption, not a coefficient taken from one validated prediction model.
Baseline probability is converted to odds. The model then applies the pooled odds ratio of 1.229 for every one-point increase in A1C reported by Zhou and colleagues.
adjusted odds = baseline odds × 1.229^(effective A1C − 7.0)risk = adjusted odds ÷ (1 + adjusted odds)With at least three A1Cs, Ampulert calculates average real variability—the average absolute change between consecutive measurements. Published research supports variability as an independent limb-risk signal; Ampulert deliberately uses smaller, custom multipliers instead of importing the study's largest hazard ratio.
| Average real variability | Ampulert multiplier |
|---|---|
| Below 0.33 | 1.00 |
| 0.33–0.519 | 1.06 |
| 0.52–0.784 | 1.13 |
| 0.785 or higher | 1.25 |
Direction-aware safeguard: a sustained improvement is not treated as harmful instability. When A1C falls by at least 0.5 points overall without a rebound of 0.5 points or more, the variability multiplier remains 1.00. Earlier high values still influence the recency-weighted A1C and fade gradually rather than disappearing immediately.
The calculated estimate is bounded between 0.3% and 18% to limit extreme extrapolation. The displayed range starts wider with one measurement and narrows modestly through six. It is a product uncertainty range—not a formal statistical confidence interval.
Consequence visualization
Worsening estimated risk produces a more severe consequence image. These bands are motivational product logic; they do not predict the anatomical level of a future procedure.
Worked example
Approximately three months between A1C measurements, with the newest measurement entered now.
Exact dates shift the estimate slightly, but measurements spaced a few months apart remain in the same image band under the current model.
Principal evidence
A 16-study meta-analysis reported an odds ratio of 1.229 for lower-extremity amputation for each one-point increase in A1C. This is the model's principal published A1C coefficient.
View publication ↗In 56,872 patients with type 2 diabetes, higher visit-to-visit A1C variability was independently associated with major adverse limb events and amputation. Its average-real-variability metric counts rises and falls equally and was calculated across a four-year history averaging 12.6 measurements. The study does not establish that a deliberate therapeutic improvement itself causes harm. Ampulert therefore uses its variability concept and thresholds, but not its 3.05 highest-quartile amputation hazard ratio, and adds a direction-aware improvement safeguard.
View publication ↗A large cohort of patients with diabetes demonstrated a graded association between A1C and lower-extremity amputation risk, supporting the model's overall direction.
View publication ↗The ADAG study established the relationship used to translate average glucose into an A1C proxy when the user has no A1C measurement.
View publication ↗Ampulert-designed assumptions
Limitations
Glycemia alone cannot fully estimate an individual's amputation risk. The current MVP does not yet incorporate several major clinical predictors:
Before stronger clinical claims are made, the model should be tested in a longitudinal U.S. dataset, evaluated for discrimination and calibration, assessed across prespecified subgroups, recalibrated, and then studied prospectively. Every result remains tied to a method version so future updates can be audited.
Transparency is part of the product