Identifying Patients Requiring Necessary Treatment in Mild to Moderate COPD

<bold>Rationale:</bold> In chronic obstructive pulmonary disease (COPD) patients identified via community surveys, treatment is necessary not only for those with poor lung function and severe symptoms but also for individuals prone to rapid disease progression. <bold>Objective:</bold> We aimed to develop a tool capable of identifying COPD patients who require necessary treatment at the initial assessment. <bold>Method:</bold> We included patients with mild to moderate COPD who received a placebo for 2 years. The natural disease progression (experiencing clinical important deterioration [CID]) without intervention of these patients were observed. The placebo groups from the 2 RCTs were utilized as the training and validation sets, respectively, for the machine learning model to predict patients requiring necessary treatment based on their baseline conditions. <bold>Result:</bold> In the placebo group of 644 patients, 248 (38.5%) with a relatively severe baseline. Despite having a better baseline, 113 patients (17.5%) developed CID during follow-up. Overall, 361 patients (56.1%) were identified as requiring necessary treatment. Among the three machine learning models, the Random Forest model had the highest Balanced Accuracy 0.852, Specificity 0.950, and PPV 0.963 in the validation set. <bold>Conclusion:</bold> Our research offers a predictive tool for identifying which mild to moderate COPD patients require necessary treatment. Table 1. Various model performances in the validation data sets <table-wrap><object-id>erj;64/suppl_68/PA1174/TB1</object-id><object-id>T1</object-id><object-id>TB1</object-id><table><colgroup><col></col><col></col><col></col><col></col></colgroup><tbody><tr><td>Statistics</td><td>Naïve Bayes</td><td>Quadratic Discriminant</td><td>Random Forest</td></tr><tr><td>Accuracy (95% CI)</td><td>0.796 (0.748, 0.838)</td><td>0.780 (0.731, 0.824)</td><td>0.827 (0.781, 0.866)</td></tr><tr><td>Sensitivity</td><td>0.819</td><td>0.809</td><td>0.755</td></tr><tr><td>Specificity</td><td>0.756</td><td>0.731</td><td>0.950</td></tr><tr><td>Positive Predictive Value</td><td>0.852</td><td>0.838</td><td>0.963</td></tr><tr><td>Negative Predictive Value</td><td>0.756</td><td>0.691</td><td>0.693</td></tr></tbody></table></table-wrap>