We conducted a multicenter, prospective, observational cohort study across 224 primary care settings in Spain between November 1, 2023, and December 1, 2024 (investigators, Appendix A). Patients were assessed at baseline and at a follow-up visit approximately 3 months later to evaluate the longitudinal properties of the RADAR score. Physicians consecutively recruited patients with confirmed COPD receiving maintenance triple therapy (ICS/LABA/LAMA) as part of routine clinical care.

Inclusion criteria were age ≥40 years, postbronchodilator FEV1/FVC <0.7 confirmed by spirometry, ≥10 pack-years smoking history, and clinical stability at baseline according to clinical consensus and major guidelines [2,3]. Exclusion criteria included other chronic respiratory diseases, inability to understand study procedures or complete questionnaires, or participation in another clinical study.

Data were collected at baseline and at follow-up (60–120-day window). A standardized case report form captured clinical variables through direct assessment, validated questionnaires, and electronic health records. Baseline treatment was categorized according to the number of inhalation devices delivering triple therapy (ICS/LABA/LAMA) as single-inhaler triple therapy (SITT), double-inhaler triple therapy (DITT), or multiple-inhaler triple therapy (MITT), defined as the use of ≥3 separate devices.

Statistical analyses were performed using R version 4.2.1. A 2-sided P<.05 was considered statistically significant. Analyses followed a complete-case approach: the full cohort (N=706) was used for baseline and outcome analyses, whereas analyses of changes in health status included patients with complete paired data (N=676). Baseline characteristics were summarized using descriptive statistics (mean [SD] or no. [%]) and compared between RADAR control groups using Welch t test or Pearson χ2 test, as appropriate. A multivariable logistic regression model was used to identify independent baseline predictors of poor control (RADAR ≥4), excluding variables included in the RADAR score. Attrition analysis compared baseline characteristics of the analytical cohort and those lost to follow-up (Appendix B, Table A2).

For study objectives, (1) prognostic utility was assessed by analyzing the association between baseline RADAR status (≥4 vs <4) and 3-month outcomes using logistic regression for binary occurrence (OR,) and negative binomial regression for event counts (incidence rate ratio,), including the natural logarithm of follow-up time as an offset; and (2) changes in control and health status were described using ΔCAT and ΔRADAR categories stratified by baseline phenotype and risk group. Supplementary multivariable analyses and model diagnostics are provided in Appendix C.

The study adhered to Good Clinical Practice principles. The protocol was approved by the Institutional Review Board (Ethics Committee for Clinical Research; Ref. 23/549-E). All participants provided written informed consent. Patient data were anonymized before statistical processing.

The patient selection process is detailed in Fig. 1. The cohort was recruited across 224 primary care centers. The final analytical cohort included 706 patients, with a median follow-up of 91 days. Table 1 presents baseline characteristics stratified by RADAR control status.

Fig. 1.

Flowchart of patient screening and inclusion. COPD, chronic obstructive pulmonary disease.

Appendix D provides definitions and risk stratification of RADAR control categories. The distribution of RADAR scores across the cohort is shown in Supplementary Fig. 1. Patients classified as poorly controlled (RADAR ≥4) exhibited a distinct clinical profile, including a higher comorbidity burden and a lower prevalence of preserved lung function vs well-controlled patients.

The poorly controlled group also demonstrated a higher symptom burden (CAT score) and poorer treatment adherence (lower TAI scores and a higher proportion with TAI ≤45). A clear monotonic relationship was observed between increasing RADAR and CAT scores (Supplementary Fig. 2). The GesEPOC phenotype distribution differed significantly, with a higher proportion of both eosinophilic and noneosinophilic exacerbators in the poorly controlled group.

Multivariable logistic regression identified independent clinical factors associated with poor control (Table 2). After adjustment, severe airflow limitation (FEV1 <50%), active smoking, higher baseline symptom burden (CAT score), and poorer adherence (TAI score) were significantly associated with RADAR ≥4. Age, sex, and comorbidity burden (Charlson index) were not significantly associated.

Fig. 2 shows the association between baseline RADAR status and 3-month outcomes. A strong association was observed for severe exacerbations (hospitalization or emergency care), which were more than threefold higher in poorly controlled vs well-controlled patients (16.2% vs 4.9%; P<.001), corresponding to an increased risk (OR, 3.75). Analysis across RADAR score gradations confirmed a progressive risk gradient for severe exacerbations (Appendix E, Table A7). Validation using alternative thresholds (≥3, ≥4, ≥5) is presented in Appendix G (Supplementary Fig. 3 and Table A9). Treating RADAR as a continuous variable confirmed this association (OR, 1.42; 95% CI, 1.28–1.58; P<.001) (Appendix H, Table A10).

Fig. 2.

Association between baseline RADAR control and 3-month adverse outcomes. Note: Panel A shows absolute event rates (%) with P values from the Pearson χ2 test. Panel B shows ORs with 95% CIs on a logarithmic scale for the association between poor control (RADAR ≥4 vs <4) and each outcome. Red bars indicate well-controlled patients (RADAR <4), and blue bars indicate poorly controlled patients (RADAR ≥4).

In contrast, results for moderate exacerbations differed by analytical approach. No association was observed when analyzed as a binary outcome (18.0% vs 18.9%; P=.520; OR, 0.94). However, when analyzed as recurrent-event burden using negative binomial regression, RADAR score was a significant predictor. Each 1-point increase in RADAR score was associated with a higher exacerbation burden (IRR, 1.20; 95% CI, 1.14–1.25; P<.001). This association remained consistent in the rate model with follow-up offset (IRR, 1.21; 95% CI, 1.15–1.27; P<.001) (Appendix F, Table A8). The composite endpoint was also more frequent in the poorly controlled group (33.8% vs 22.9%; P=.001; OR, 1.72), driven primarily by severe exacerbations.

Importantly, the association between poor RADAR control and severe exacerbations remained robust after multivariable adjustment. We constructed a logistic regression model adjusting for age, sex, FEV1 <50%, active smoking, and baseline CAT score. As shown in Table 3, poor clinical control (RADAR ≥4) was the strongest independent predictor of severe exacerbations (aOR 2.88; 95% CI 1.62–5.12; P<0.001), carrying nearly triple the risk after accounting for lung function and smoking status.

Fig. 3 shows the 3-month distributions of ΔCAT and ΔRADAR for 676 patients with complete data. Patients were stratified by baseline risk (high vs low), with the high-risk group further subdivided by phenotype.

Fig. 3.

Distribution of 3-month changes in health status (ΔCAT) and control score (ΔRADAR). Note: Panel A shows 3-month change in COPD Assessment Test score (ΔCAT), and Panel B shows 3-month change in RADAR score (ΔRADAR), with patients in both panels stratified by baseline risk phenotype. Categories for both scores were defined as improved (Δ ≤−2), stable (Δ=−1 to +1), and worsened (Δ ≥+2).

Panel A (ΔCAT) shows a clear association between baseline risk and health status changes. A higher proportion of high-risk patients showed improvement compared with low-risk patients (e.g., 67.1% in high-risk noneosinophilic exacerbators and 57.1% in high-risk nonexacerbators vs 31.0% in the low-risk group).

Panel B (ΔRADAR) showed a similar pattern, with greater improvement among high-risk groups (e.g., 71.0% in high-risk noneosinophilic exacerbators vs 14.0% in the low-risk group).

The RADAR score demonstrated strong predictive capacity for severe exacerbations (Table 3). Patients with RADAR ≥4 had a significantly higher incidence of severe events, highlighting its clinical utility in identifying patients requiring closer monitoring or proactive intervention [17].

Real-world data show that many patients receiving maintenance therapy remain symptomatic and experience exacerbations [18–20]. Optimizing treatment during follow-up remains essential. However, clinical control is frequently not assessed, and therapeutic adjustments are often not implemented in poorly controlled patients [6]. This therapeutic inertia contributes to increased morbidity and mortality [21,22]. This inertia may result from underestimation of disease severity by clinicians and a discrepancy between patient and physician perceptions. Many patients normalize symptoms or perceive adequate control despite significant limitations [23–25]. The RADAR score provides an objective assessment tool to bridge this gap and facilitate proactive management.

The predictive signal for moderate exacerbations depended on the analytical approach. While no association was observed for binary outcomes, RADAR score was associated with recurrent-event burden. This suggests that RADAR captures disease instability reflected by event frequency rather than isolated episodes. This aligns with previous evidence establishing RADAR ≥4 as a threshold for adverse outcomes [11].

Our findings indicate the RADAR score retains the short-term risk-predictive capacity of the original, validated COPD clinical control questionnaire [26–30]. While the BODE index is a powerful predictor of long-term survival, its reliance on the 6-min walk distance limits its feasibility in primary care. In contrast, the RADAR score captures similar dimensions (dyspnea, activity) using parameters readily available in routine visits, making it more suitable for monitoring short-term fluctuations in stability. Moreover, unlike unidimensional health status instruments (e.g., CAT or CCQ) [14,26], the RADAR score integrates clinical history (exacerbations, medication) to specifically stratify the risk of imminent severe events. However, as a quantitative measure derived from weighted qualitative variables [4,5], the RADAR score offers advantages: greater discrimination of control levels and the ability to quantify changes post-intervention. Its variables (dyspnea, physical activity, rescue use, exacerbation history) are easily obtained in routine practice, facilitating its implementation as a decision-making tool for monitoring and adjusting treatment, as recommended by guidelines.

Importantly, RADAR ≥4 identifies a clinically meaningful subgroup characterized by severe airflow limitation, active smoking, higher symptom burden, and poor adherence, all of which are established predictors of adverse outcomes [31,32]. The RADAR score therefore functions as an effective screening tool for identifying high-risk patients in primary care.

Although 60% of the cohort was classified as controlled at baseline, control status changed frequently during the 3-month follow-up. This dynamic pattern, together with the finding that RADAR ≥4 predicts short-term severe exacerbations, confirms that clinical control is a useful parameter for visit-to-visit decision-making. In this cohort, status changed in more than 50% of high-risk patients, particularly frequent exacerbators, but in only slightly more than 30% of low-risk patients. Soler-Cataluña et al. [6] similarly found that one-third of patients with COPD changed control status over 3 months, with approximately half improving and half worsening. Changes in control status appear to occur more frequently than changes in GOLD stage, phenotype, or risk level, indicating that these measures assess different dimensions of disease [6,30].

This expected variability, given that RADAR includes labile variables such as dyspnea, rescue medication use, physical activity, and exacerbations, was evident in our analysis. Significant 3-month variation was observed in both ΔCAT and ΔRADAR, which showed remarkably similar distribution patterns according to baseline phenotype and risk level. A significantly larger proportion of patients in the high-risk exacerbator phenotypes reported improvement compared with the low-risk group.

This study was not designed to identify factors associated with short-term changes in control. When interpreting this pattern, it is important to consider that baseline status influences subsequent changes [31–33], as does the potential effect of clinical interventions, which are more likely to be implemented in high-risk patients. Because of the observational design, the impact of such interventions or medication changes was not assessed.

Analysis of the cohort's clinical characteristics is informative: 70% met GesEPOC high-risk criteria, and nearly half were exacerbators. In addition, treatment complexity was greater among poorly controlled patients. Use of multiple-inhaler triple therapy (≥3 devices) was significantly more frequent in poorly controlled patients than in those receiving single-inhaler triple therapy (Appendix I, Table A11). Poor adherence, measured by TAI, was identified as a key modifiable factor in nearly half of the cohort and was significantly more frequent in poorly controlled than in well-controlled patients (56.2% vs 40.5%; P<.001).

Treatment complexity is a known modifiable risk factor, and simplification of the regimen or reduction in the number of inhaler devices may improve adherence and clinical outcomes. Single-inhaler vs multiple-inhaler triple therapy has been associated with improved persistence, better adherence, and lower exacerbation rates [34–37]. The EPOCONSUL audit identified switching molecules and inhaler devices as the most frequent intervention during pulmonology follow-up visits [9].

A major strength of this study is its prospective, multicenter design in a large, real-world primary care cohort of patients receiving maintenance triple therapy, which enhances generalizability to this treated high-risk subgroup. Standardized data collection and use of validated instruments also strengthen the findings.

Several limitations should be acknowledged. First, the observational design precludes causal inference. Second, the 3-month follow-up is too short to assess long-term outcomes such as mortality. Third, although the sample size was determined by recruitment feasibility, post hoc sensitivity analysis confirmed that the final cohort provided sufficient statistical power to detect clinically relevant associations for the primary outcomes (Appendix B). Nevertheless, unobserved outcome events may have been more frequent among patients lost to follow-up, and this possibility cannot be completely excluded; however, if patients lost to follow-up experienced higher event rates, the reported risk estimates would likely be conservative. Fourth, the cohort consisted exclusively of patients receiving maintenance triple therapy, with more than 70% classified as high risk according to GesEPOC criteria. Therefore, these results should not be generalized to patients with milder disease or to those receiving mono- or dual-bronchodilator regimens. Finally, residual confounding, inherent to observational research, remains possible.

These findings have direct implications for primary care. The RADAR score is a simple tool that identifies patients at significantly increased short-term risk of severe exacerbations. A high baseline RADAR score should be considered an immediate clinical alert prompting comprehensive assessment, including adherence review and comorbidity management, to mitigate this elevated risk [2,3]. Thus, RADAR may serve as a practical risk-stratification tool in routine care.

The observed changes in CAT and RADAR scores underscore the dynamic nature of COPD and reinforce the need for regular follow-up and reassessment [2,3]. The causes of these changes were not investigated. An uncontrolled status should be interpreted as a signal to act, although it does not in itself identify the reason for poor control or the specific intervention required. However, beyond estimating short-term prognosis, the RADAR score may help clinicians identify the variables contributing to poor control at each visit.

Future research should validate these findings over longer follow-up periods (≥1 year) to assess sustained risk and mortality. Interventional studies are also needed. A randomized trial comparing standard care with a RADAR-guided strategy, in which a high score triggers predefined interventions, would determine whether this proactive approach improves outcomes such as hospitalization rates.