Predictive model identifies women with COPD in need of palliative care

A newly developed prediction model has demonstrated robust performance in recognizing women with chronic obstructive pulmonary disease (COPD) who might benefit from proactive palliative care. However, external validation is warranted.

“[W]e developed a predictive model for 1-year all-cause mortality among women with COPD using data from the older Australian Longitudinal Study on Women’s Health (ALSWH) cohort,” the researchers said.

The final model included six key predictors, namely smoking status, underweight BMI, regular need for assistance with daily activity, prescription of four or more medications, duration of illness, and number of hospital admissions. [Respirology 2025;30:623-632]

The model exhibited good predictive performance, with an area under the receiver operating characteristic (AUROC) of 0.82 (95 percent confidence interval, 0.80–0.85), and excellent calibration. Using a cutoff of 56.6 percent predicted risk, the model had a sensitivity of 72.3 percent, specificity of 77.7 percent, and accuracy of 75.0 percent.

In the decision curve analysis (DCA), the predictive model offered a higher net benefit for clinical decision-making.

“This model can be used to stratify women with COPD into different risk groups, enabling targeted palliative care intervention,” the researchers said. “For patients with similar population characteristics, this model can support healthcare providers in identifying high-risk individuals and tailoring interventions accordingly.” 

However, external validation is required to validate its performance and applicability across different clinical settings. Such validation will also improve the model’s generalizability.

Predictors

The six prognostic determinants used in this model were independently associated with a high risk of mortality. These predictors identify important aspects of disease severity and progression, such as ongoing respiratory damage and susceptibility to exacerbations from smoking, as well as nutritional depletion from underweight BMI. [BMJ Supportive & Palliative Care 2024;14:e2316-e2329]

Other aspects captured included multimorbidity, treatment burden and complexity from multiple prescribed medications; chronic disease progression from longer illness duration; and severe exacerbations and health deteriorations from frequent hospital admissions. [Diagnostics 2023;13:1344]

In addition, the model takes into account a more severe disease progression where functional decline can substantially impact daily living. Incorporating all these variables allows the model to do a comprehensive evaluation of patients, identifying those at high risk who may benefit from timely interventions and referral to palliative care in the last year of life.

"These results indicate that the predicted mortality probabilities of the model [are] closely aligned with the observed mortality rates, highlighting its robust accuracy,” the researchers said. “In contrast, a statistically significant (p<0.05) result suggests discrepancies between the predicted and actual outcomes, implying a poorer calibration.

Participants

A total 1,236 diagnosed with COPD from the 1921–1926 ALSWH cohort were included in the analysis. Lasso regression and logistic regression were used to identify predictors.

The researchers assessed the predictive performance of the model using the AUROC curve, calibration plot, and calibration metrics. They established the optimal cutoff point for risk classification using the Youden index and evaluated the clinical utility of the model using DCA.