PerioRisk gum disease risk check

How likely is gum disease?

Answer 14 short questions to estimate your chance of moderate or severe gum disease (periodontitis). The questions come from a US national health survey in which 10,660 adults also had a full gum examination by a dentist.

Your answers never leave this device Tested on 3,610 different people Open source & free

Before you start

About you

Your mouth

For the yes/no questions, choose “Not sure” if you don’t know — the estimate then uses a typical answer instead.

How accurate is it?

The model learned from 7,050 adults in NHANES 2009–2012 and was then tested, once, on 3,610 different adults from NHANES 2013–2014 — a later survey it had never seen. Every one of them had a full-mouth gum examination, scored with the CDC/AAP definition used for national surveillance. No estimate is a diagnosis: about 17 in 100 people in the lower-risk band did have moderate or severe gum disease.

0.78AUC — 0.5 is a coin toss, 1.0 is perfect
87%of people with moderate or severe gum disease got an estimate of 30% or more
80%of people without it got an estimate below 60%
3,610people in the independent test survey
Calibration plot: predicted probability against the observed proportion with periodontitis in the 2013–2014 test set; points lie close to the diagonal, slightly below it between 40% and 80%.
Calibration: do the percentages match reality?
ROC curves of logistic regression, random forest, gradient boosting and the final calibrated model on the 2013–2014 test set; all have an AUC of about 0.78.
Discrimination: three model types perform almost identically
Decision curve: the model gives more net benefit than referring everyone or no one for threshold probabilities between about 10% and 80%.
Usefulness: net benefit compared with referring everyone

Limits you should know

  • It was built on adults in the United States. How well it works in other countries is not yet known.
  • It estimates whether gum disease is present now, not your future risk.
  • On average it slightly over-estimates: in the test survey, for every 100 cases it expected, about 90 were found (about 80 for the US adult population as a whole). Treat the percentage as a guide, and the risk band as the message.
  • It is less accurate for people aged 65 and over, people with diabetes and people who did not finish high school — groups that are also at higher risk.
  • Only the original English survey questions were tested; translated versions have not been validated.
  • It has not been tested in dental clinics or followed over time. It is not a medical device.

About PerioRisk

Why it exists

About four in ten US adults aged 30 and over who have teeth have periodontitis, and it is often painless until late. A full gum examination needs a dental visit; a short, validated questionnaire can help people decide to have one and helps researchers estimate gum disease in surveys without examinations.

How it was built

Open NHANES data (CDC/NCHS, public domain) → the CDC/AAP case definition applied to every measured site between the teeth (wisdom teeth excluded) → three model types compared by cross-validation → gradient boosting with probability calibration → one test on a later survey cycle, including calibration, decision-curve, subgroup and survey-weighted analyses, reported following TRIPOD+AI. The whole pipeline re-runs with one command and reproduces the published model exactly.

Privacy

The model runs inside this page. Your answers are not uploaded, stored, logged or tracked — there are no cookies and no analytics. Closing the tab erases them.

For researchers and developers

pip install periorisk

import periorisk
r = periorisk.predict(age=58, sex="male", smoking="current",
                      gum_health="fair", loose_teeth="yes", floss_days=0)
print(r.percent, r.band)            # estimate and risk band
periorisk.recalibrate(r.probability, target_prevalence=0.30)

Source code, results report, model card and TRIPOD+AI checklist: github.com/Hezarani/periorisk. To cite PerioRisk: Boustani Hezarani H. PerioRisk: periodontitis risk screening from self-reported questions (software). Zenodo; 2026. doi:10.5281/zenodo.23813178.

Author

Hossein Boustani Hezarani, DDS · ORCID 0009-0007-8481-0498 · part of the open DentGit dental tools.

0 of 14 answered