Tool predicts new-onset exertional, resting hypoxemia in patients with fibrotic ILD

Key takeaways:

• A validated hypoxemia prediction tool includes variables that are “readily available” to clinicians.
• The tool had “good” discrimination and calibration in predicting exertional and resting hypoxemia.

A tool that includes age, BMI, two lung function measures and a diagnosis of idiopathic pulmonary fibrosis can predict new-onset exertional and resting hypoxemia in fibrotic interstitial lung disease, according to study results.

These data were published in Annals of the American Thoracic Society.

“This validated clinical prediction tool allows timely identification of patients at increased risk of hypoxemia in day-to-day patient care,” Yet H. Khor, PhD, clinical senior lecturer at The University of Melbourne, told Healio. “The prediction tool is an easy-to-use score-based format based on readily available parameters in regular clinical assessments.

“The prediction tool has been developed using patient populations of different backgrounds, ILD diagnoses and disease severity, so it is widely applicable to different populations,” Khor continued.

Using multiple cohorts of patients with fibrotic ILD, Khor and colleagues sought to construct and validate a risk prediction tool that could foresee new-onset exertional and resting hypoxemia at 6 months in this patient population.

As Healio previously reported, patients with IPF vs. non-IPF fibrotic ILD appear to have higher cumulative incidences of exertional and resting hypoxemia.

Researchers derived and tested the tool using the Canadian Registry for Pulmonary Fibrosis cohort (n = 2,515), whereas validation was done in three different cohorts (total, n = 508), including two from Australia (Alfred Health and Austin Health) and one from the University of California at Davis in the U.S.

A measure of nadir oxyhemoglobin saturation less than 88% during the 6-minute walk test was used to classify new-onset exertional hypoxemia, and resting oxyhemoglobin saturation less than 88% was used to classify new-onset resting hypoxemia, according to researchers. Ambulatory/continuous oxygen use was also used to classify hypoxemia in this study.

Researchers assessed several candidate predictor variables for use in the risk prediction tool: age, sex, smoking status, BMI, ILD subtype, 6-minute walk distance and various measures of lung function.

After finding the best-performing model through time-varying Cox regression analysis, researchers used Harrell’s C-index to evaluate discrimination (“good” = value ≥ 0.7) and the good-of-fit (GoF) likelihood ratio test to evaluate calibration (“good” = P ≥ .05).

Presence of hypoxemia

Of the 1,969 patients in the derivation cohort evaluated for new-onset exertional hypoxemia, 487 patients (25%) developed this type of hypoxemia during follow-up (median, 2.4 years). This was also the case for 156 of 413 patients (38%) in the validation cohort (median follow-up, 2.3 years).

In terms of resting hypoxemia, 232 (11%) of the 2,186 patients in the derivation cohort and 87 (18%) of the 495 patients in the validation cohort assessed for resting hypoxemia had this type of hypoxemia during follow-up.

Model performance

Researchers found that the top-performing prediction models (considering both Akaike information criterion and clinical usability) for both hypoxemia types factored in age, BMI, an IPF diagnosis, percent-predicted FVC and percent-predicted diffusing capacity of carbon monoxide (DLCO).

“Predictors for new-onset exertional and resting hypoxemia in patients with fibrotic ILD were consistent with those identified for prognostication,” Khor said. “While this was not totally unexpected, this highlights the significance of hypoxemia in this population.”

When tested in the derivation cohort, researchers noted “good performance” of the tool for predicting both exertional hypoxemia (C-index, 0.7; GoF, P = .85) and resting hypoxemia (C-index, 0.77; GoF, P = .27).

In the validation cohort, the tool performed well in predicting exertional hypoxemia (C-index, 0.72; GoF, P = 1). Notably, calibration of the tool for predicting resting hypoxemia was “suboptimal” (GoF, P = .001), whereas discrimination was good (C-index, 0.78).

“Suboptimal GoF in the validation cohort likely reflected overestimation of hypoxemia risk and indicated that the model is not flawed because of underestimation of hypoxemia,” Khor and colleagues wrote.

As an additional measure of the tool, researchers divided patients based on their risk level for developing exertional and resting hypoxemia at 6 months using the score (0-11) derived from the tool and captured incidence of both hypoxemia types at baseline and the last reported follow-up.

At baseline, more high-risk patients (score > 5) vs. low- (score 0-3) and moderate-risk (score 4-5) patients had exertional hypoxemia (19% vs. 3.8% vs. 8.3%), and this was also true at the last follow-up (46.3% vs. 12.2% vs. 28.1%).

For resting hypoxemia, researchers also found a higher proportion of high-risk score vs. low- or moderate-risk score patients with this type of hypoxemia at baseline (7.3% vs. 0.2% vs. 2.6%). This same pattern was found when assessing incidence at the last follow-up (24.3% vs. 1.4% vs. 9.4%).

“This prediction tool can help to better select suitable patients for participation in clinical trials, particularly for the evaluation of supplemental oxygen therapy and provision of supportive care,” Khor told Healio.