Frontiers in Artificial Intelligence

Models only modestly identify airflow obstruction in older Indians

A combined model scored 0.686 for distinguishing people with and without obstruction internally and 0.665 in later validation; the authors called its performance modest.

Paper by Rupeshkumar Balasaheb Deshmukh, Sudhanshu Mahajan · Bharati Vidyapeeth (Deemed to be University)
Figure 1. Participant flow from the full LASI wave 1 sample.
Open the figure at full size
Figure 1. Participant flow from the full LASI wave 1 sample.Participant flow from the full LASI wave 1 sample.Rupeshkumar Deshmukh et al., CC BY

Our read

Biomarkers
Studied:
Cross-sectional study of 35,051 older Indian adults with usable spirometry
What changed:
Models identified airflow obstruction only modestly, with leading models reaching an AUC of about 0.65–0.69
The catch:
The fixed-ratio definition over-identifies obstruction in older adults, partly rewarding models for predicting age

Editor of record: Boris Djordjevic · AgingResearch.News

Follow @agingbrief

In an observational analysis of 35,051 older Indian adults, researchers tested whether routine survey information could identify airflow obstruction, or difficulty moving air out of the lungs. They compared predictions with spirometry, a test that measures how much air people exhale and how quickly.

Obstruction was present in 4,212 participants (12.0%). Only 10.9% of those with obstruction had ever reported a chronic lung disease diagnosis. Complex machine-learning models distinguished people with and without obstruction only modestly. They showed no demonstrable advantage over a simpler statistical model. A five-item score missed about half of people with obstruction and was least sensitive in women. Referrals based on predicted risk showed no consistent advantage over referring everyone for spirometry.

Why it matters

Chronic obstructive pulmonary disease is substantially underdiagnosed in India. For older adults, the study addressed whether routine health information could help prioritize lung testing where capacity is limited.

Caveats

The analysis used one survey wave, and performance weakened under an alternative definition of obstruction. The authors called for prospective evaluation and said the score should not replace spirometry or clinical judgment.

From the paper

Participants

Among the 66,606 respondents aged 45 years and above, 35,051 entered the analytic sample.

Results, 3.11.9 Inverse-probability weighting for selection into the analytic sample

Result

… we additionally computed a survey-weighted estimate using the LASI spirometry-specific sampling weight, with a primary-sampling-unit cluster bootstrap: 12.6% (95% CI 11.8–13.5%, 2,418 primary sampling units).

Results, 3.4 Survey-weighted national prevalence

Events by group

Of the 35,051 participants in the analytic sample, 28,805 (82.2%) had a graph flagged acceptable and 6,246 (17.8%) were flagged usable only; 27,945 of the acceptable group were also flagged repeatable.

Results, 3.14.1 Restriction to spirometry flagged acceptable

Limitation

First, the GOLD fixed-ratio definition, applied before bronchodilator, is known to over-identify obstruction in older adults relative to lower-limit-of-normal definitions.

4.9 Limitations

Quoted word for word from the full text on frontiersin.org.

Comments

Be the first to comment.