65/100

Moderate evidence

Cite
More
EmailLinkedInX

AI scorecard shows 15.9-fold pneumonia risk gap in older Japanese, yet needs more testing

Study title: A Machine Learning-Derived Risk Scorecard for Pneumonia Hospitalization in Japanese Old-Old Adults

Evidence type
Observational study
Sample
n = 1,098,404
Sources
1 sourcePeer-reviewed
Published
1 min read

Study title: A Machine Learning-Derived Risk Scorecard for Pneumonia Hospitalization in Japanese Old-Old Adults

Researchers developed and internally validated a machine learning-based risk scorecard for 1-year pneumonia hospitalization among 1,098,404 community-dwelling Japanese adults aged 75 years or older who completed the QMCOO frailty screening between April 2020 and March 2024. Using Late-Stage Medical Care System claims, a 20-learner Super Learner ensemble and a 17-feature 0-21 point scorecard were trained and tested, with the scorecard achieving AUC 0.786 and good calibration.

Read the study abstract

Aim To develop and internally validate a machine learning-based risk scorecard for 1-year pneumonia hospitalization among community-dwelling Japanese adults aged ≥ 75 years using routinely collected frailty screening and claims data. Methods We conducted a retrospective cohort study of 1 098 404 community-dwelling adults aged ≥ 75 years who completed the Questionnaire for Medical Checkup of Old-Old (QMCOO) between April 2020 and March 2024, using the Late-Stage Medical Care System claims database. Data were spli…

A Machine Learning-Derived Risk Scorecard for Pneumonia Hospitalization in Japanese Old-Old Adults
Photo by engin akyurt on Unsplash · https://unsplash.com/@enginakyurt?utm_source=truvace&utm_medium=referral

The gain

On internal test data, a 17-feature AI-derived scorecard sorted Japanese adults aged 75 and over into groups whose 1-year pneumonia hospitalization rates were 0.12%, 0.49% and 1.91%, a 15.9-fold gradient.

In brief

The tool stratified the population into low, moderate, and high risk groups with a 15.9-fold gradient in observed hospitalization rates, suggesting potential utility for identifying high-risk individuals during routine checkups. As of the October 2026 publication date, the findings reflect internal validation only, and the authors note that external validation would be needed before supporting targeted preventive assessment in primary care.

Main points

  1. Retrospective cohort of 1,098,404 community-dwelling adults aged ≥ 75 years who completed the Questionnaire for Medical Checkup of Old-Old between April 2020 and March 2024
  2. Data source was the Late-Stage Medical Care System claims database with 70% training and 30% test split
  3. Outcome was 1-year pneumonia hospitalization defined by ICD-10 J12-J18, J69, occurring in 4525 participants (0.41%)
  4. 17-feature point-based scorecard (0-21 points) derived from training set with Platt calibration

The full story

The study used routinely collected frailty screening via the Questionnaire for Medical Checkup of Old-Old and claims from the Late-Stage Medical Care System. Participants had mean age 80.6 years and 40.3% were male. The prediction target was hospitalization with ICD-10 codes J12-J18 and J69 within one year.

Performance was reported on a held-out 30% test set. The full ensemble reached AUC 0.823 and the simplified scorecard reached AUC 0.786 with calibration slope 1.017 and calibration-in-the-large -0.001. Risk stratification showed event rates of 0.12% in low, 0.49% in moderate, and 1.91% in high groups.

Sources

1 on file
  1. Peer-reviewed2026-10-01

Talk it over

Three questions to start on, two more if you have time
The scorecard sorted adults aged 75 and over into groups with 1-year pneumonia hospitalization rates of 0.12% and 1.91%. What would a clinic or health office do differently for the high-risk group, and who should decide that?

Start with a question

AI scorecard shows 15.9-fold pneumonia risk gap in older Japanese, yet needs more testinghttps://truvace.com/item/a-machine-learning-derived-risk-scorecard-for-pneumonia-hospitalization-in-japan

  1. The scorecard sorted adults aged 75 and over into groups with 1-year pneumonia hospitalization rates of 0.12% and 1.91%. What would a clinic or health office do differently for the high-risk group, and who should decide that?

    Answer in the thread
  2. The researchers tested the scorecard on a held-out 30% of the same 1,098,404 people who completed the frailty screening. How much does a test on the same population tell us about other prefectures or health systems, and what would settle it?

    Answer in the thread
  3. The study drew on people who completed the Questionnaire for Medical Checkup of Old-Old between April 2020 and March 2024. Who might be missing from that group, and how could that change the scorecard's results?

    Answer in the thread
If you have more time
  • The full ensemble reached AUC 0.823 and the 17-feature scorecard reached AUC 0.786. How would you weigh a small drop in performance against a tool a checkup nurse could use by adding up points?

    Answer in the thread
  • The authors say external validation is needed. What kind of outside data, from which places or groups of older adults, would settle whether the scorecard holds up?

    Answer in the thread

What readers say