Pulmonology · Medical Literature
Classification and regression trees to identify COPD subgroups in clinical trial populations: insights from the IMPACT trial.
Abstract
Heterogeneity in chronic obstructive pulmonary disease (COPD) challenges the identification of optimal treatment populations. Subgroups derived from real-life cohorts using classification and regression trees (CARTs) have shown prognostic value for mortality, but their applicability to clinical trial populations to predict treatment response needs to be further investigated. We sought to evaluate whether previously identified CART-based subgroups are relevant for predicting outcomes and treatment response in the IMPACT trial, and to assess the stability of these subgroups using de novo clustering. Post-hoc analysis of the IMPACT trial, a randomized controlled trial comparing inhaled corticosteroid (ICS)-containing dual or triple therapy to dual long-acting bronchodilation in patients with COPD. CART-based classification from prior real-life cohorts was applied to trial participants. Additionally, de novo clustering was performed using factor analysis of mixed data to identify alternative subgroup structures. Outcomes included mortality, exacerbation rates, lung function, dyspnea, and health status. Subgroups were compared descriptively in terms of baseline characteristics and on-treatment outcomes, with particular attention to blood eosinophil count and treatment response. Among 10,355 patients, both CART-based (5 classes) and de novo (5 clusters) classifications identified subgroups with distinct baseline profiles and variable on-treatment outcomes. Exacerbation and mortality rates differed markedly across subgroups, with numerically greater differences between treatment arms in higher-risk groups. Most clusters showed heterogeneous patterns of outcomes, while one cluster, characterized by elevated blood eosinophils (median 930/mm³), showed a numerically lower exacerbation rate with triple versus dual bronchodilation. Improvements in lung function and symptom scores were also more pronounced in this group. Despite limited concordance between the two methods, both consistently identified subgroups with higher event rates and greater numerical separation between treatment arms, supporting their potential value for clinical trial enrichment and personalized treatment strategies. Application of a CART-based classification derived from real-life cohorts to a clinical trial population revealed subgroups with distinct baseline characteristics and differential treatment outcomes. These findings should be interpreted as exploratory and hypothesis-generating, and may inform future work on trial enrichment strategies and personalized approaches to COPD management. The IMPACT trial was registered on ClinicalTrials.gov under number NCT02164513, first submitted on 12 June 2014.
Clinical relevance
Can CART-derived COPD subgroups, originally developed and validated for mortality prediction in real-life cohorts, be applied to a clinical trial population (IMPACT) to meaningfully predict differential treatment response to ICS-containing versus ICS-free dual bronchodilation? __SNAPSHOT__ {"studyDesign":"Post-hoc secondary analysis of a phase 3 randomized controlled trial (IMPACT), using two parallel subgrouping approaches: (1) application of externally derived CART-based classification, and (2) de novo unsupervised clustering via factor analysis of mixed data (FAMD)","setting":"Multinational; clinical trial population from the IMPACT trial (NCT02164513); academic and clinical research centers across multiple countries","sampleSize":"Not reported in abstract; the parent IMPACT trial enrolled 10,355 patients randomized 2:2:1","population":"Adults with moderate-to-very-severe COPD and at least one moderate or severe exacerbation in the prior year, as per IMPACT trial eligibility criteria","intervention":"ICS-containing therapy: single-inhaler triple therapy (fluticasone furoate/umeclidinium/vilanterol, FF/UMEC/VI) or ICS-containing dual therapy (fluticasone furoate/vilanterol, FF/VI)","comparator":"ICS-free dual long-acting bronchodilation (umeclidinium/vilanterol, UMEC/VI)","primaryOutcome":"Not formally defined as a single primary endpoint in this post-hoc analysis; outcomes evaluated descriptively across subgroups included mortality, exacerbation rates, lung function (FEV1), dyspnea, and health status (SGRQ)","followUp":"52 weeks (consistent with the parent IMPACT trial design)","fundingSource":"Not reported in the abstract; the parent IMPACT trial was industry-sponsored by GlaxoSmithKline (NCT02164513)"}
Key findings
The abstract as provided is truncated and does not report quantitative results (effect sizes, rate ratios, confidence intervals, p-values, or NNT/NNH) for any of the outcomes examined. Accordingly, no numerical findings can be cited or independently calculated without access to the full text. What the abstract confirms methodologically is: (1) CART-based subgroup classification derived from prior real-life European COPD cohorts (Initiatives-BPCO and related datasets) was applied to IMPACT trial participants; (2) de novo clustering using FAMD was performed as an independent stability check; and (3) subgroups were compared descriptively on baseline characteristics and on-treatment outcomes including mortality, exacerbation rates, lung function, dyspnea, and health status. The broader CART phenotyping framework from prior Burgel/Paillasseur publications identified clinically distinct phenotypes — including a low-comorbidity group, a severe airflow limitation/low-BMI group, and a cardiovascular comorbidity-predominant group — each with differing mortality trajectories in real-life cohorts. Whether these distinctions translated into differential treatment response to ICS-containing regimens within the IMPACT population is not ascertainable from the available abstract. All key quantitative findings are: not reported.
Practice implications
This study addresses a clinically important question: whether phenotype-guided COPD classification tools developed in observational, real-life cohorts retain predictive and treatment-discriminating value when applied to clinical trial populations enriched for exacerbation risk. The tension between real-world phenotypic heterogeneity and clinical trial eligibility criteria is well established — trial populations typically exclude the most comorbid and functionally impaired patients, compress phenotypic variance, and skew toward a higher-exacerbation-risk profile that may not map cleanly onto cohort-derived subgroups. The additional de novo FAMD clustering serves as an internal validity check on whether the imported CART structure is the most parsimonious fit for the trial data. However, because only the truncated abstract is available, it is not possible to determine: (1) whether CART-based subgroups demonstrated prognostic separation for any outcome in the IMPACT population; (2) whether ICS-benefit (exacerbation reduction, mortality benefit) was heterogeneous across subgroups; (3) the degree of concordance between imported CART classification and de novo FAMD clusters; or (4) whether any subgroup-by-treatment interaction was statistically significant or pre-specified. Clinicians should not alter prescribing practice based on this abstract alone. Findings apply to patients meeting IMPACT eligibility (moderate-to-very-severe COPD with recent exacerbation history); patients with mild COPD, no exacerbation history, or significant exclusionary comorbidities were not represented. This summary is limited to abstract-level data only. __LIMITATIONS__ Post-hoc, exploratory subgroup analysis of a trial not designed to test CART-based phenotyping; all subgroup findings are hypothesis-generating only and susceptible to type I error without pre-specified multiplicity adjustment.,The IMPACT trial population is highly selected (moderate-to-very-severe COPD, mandatory exacerbation history, extensive exclusion criteria), which compresses phenotypic heterogeneity and may prevent adequate representation of all CART-derived subgroups, particularly the low-risk 'mild' phenotype.,No formal primary endpoint or inferential statistical framework is defined for this post-hoc analysis, making interpretation of outcome differences across subgroups descriptive rather than confirmatory.,External CART classification was developed and validated in real-life European cohort populations; transportability to a global, trial-enriched population with different baseline characteristics has not been established.,Factor analysis of mixed data (FAMD) with de novo clustering is an unsupervised method susceptible to instability across sample splits, choice of hyperparameters, and the specific variable set included — the concordance between CART-imported and FAMD-derived subgroups is not reported in the abstract.,Only 52 weeks of follow-up may be insufficient to detect meaningful mortality differences across phenotypic subgroups.,Full results, statistical analyses, and data on adverse events across subgroups are not available from the abstract; all quantitative findings are not reported.,Potential industry sponsorship of the parent IMPACT trial (GSK) introduces risk of selective reporting bias, though this post-hoc analysis appears investigator-initiated. __BOTTOM_LINE_CATEGORY__ promising_but_preliminary __BOTTOM_LINE__ This post-hoc analysis of the IMPACT trial applies validated CART-based COPD phenotyping tools — previously shown to have prognostic value for mortality in real-life cohorts — to a clinical trial population to assess their utility in predicting treatment response to ICS-containing versus ICS-free dual bronchodilation. The study addresses a meaningful precision medicine question but remains exploratory and descriptive, and no quantitative results are available from the truncated abstract. Clinicians should await full publication before drawing any practice-relevant conclusions about phenotype-guided ICS prescribing.