Development and validation of a clinical nomogram for frailty prediction in older COPD patients: a machine learning approach

Abstract

Objective

Frailty is associated with adverse clinical outcomes in older patients with chronic obstructive pulmonary disease (COPD). Early prediction for frailty could be beneficial for more accurate prevention and management. This study aimed to construct a practical and convenient risk prediction model for frailty in older COPD patients.

Methods

Older COPD patients from China Elderly Comorbidity Medical Database(CECMed) were assessed using the FRAIL scale. Data included demographics, socioeconomic status, medical history, medication use, vital signs, and comprehensive geriatric assessment. Predictive variables were selected using LASSO regression. Five machine learning algorithms were employed to construct prediction models. The optimal model was determined by AUC values and evaluated through calibration curve analysis and DCA. The model was shown by a nomogram.

Results

A total of 860 older COPD patients were analyzed, and 357 (41.5%) were pre-frailty and 195 (22.7%) were frailty. LASSO regression identified six predictors: age, CCI, bronchodilator utilization, DBP, gait speed, and grip strength. The optimal model was the logistic regression model (AUC = 0.858, 95%CI 0.815–0.897, 61.9% of sensitivity and 90.2% of specificity). Validation through bootstrapping revealed excellent calibration, and DCA indicated that the model has good performance in the threshold range of 0.25 to 1.0. The nomogram was constructed based the six predictors for predicting frailty and pre-frailty.

Conclusion

This study revealed the prevalence of frailty in older COPD patients and identified low DBP as a novel risk factor for frailty. We developed a predictive model based on readily available clinical parameters, providing a practical and convenient tool for predicting frailty in older COPD patients.

Background

Chronic obstructive pulmonary disease (COPD) is a leading cause of global mortality, with a significant prevalence, mortality rate, and disease burden [1]. According to the China Lung Health Study (CPH), the prevalence of COPD in people aged 60–69 years is 21.2%, while the prevalence of COPD in people aged 70 years and over is 35.5%, indicating a substantial disease burden in this aging population [2]. Frailty is a complex syndrome associated with physiological vulnerability, reflecting not only age-related physical decline but also influencing the course of chronic illnesses. This impact is particularly notable in COPD patients, leading to decreased functional capacity and increased risks of hospitalization and mortality [3]. A meta-analysis revealed that COPD patients are nearly twice as likely to exhibit frailty compared to non-COPD individuals (odds ratio = 1.97, 95%CI 1.53–2.53) [4]. The presence of frailty in COPD patients can increase the risk of readmission, prolonged hospitalization, deteriorated quality of life, and elevated mortality [5–7]. Early identification of frailty and timely interventions, such as nutritional support and exercise guidance, may improve clinical outcomes in COPD patients with frailty [5,8].

At present, there are few attempts to develop risk prediction models for screening high-risk groups of frailty in patients with COPD. Most related studies focus on investigating the frailty status of these patients and the influencing factors. Currently, only one relevant literature report was published by Yong Chen et al. in 2024 [9]. This model utilized seven machine learning methods and created an online predictive risk modeling website using the Chinese Health and Retirement Longitudinal Study (CHARLS) database. Depression, smoking, gender, social activities, dyslipidemia, asthma, and residence type (urban vs. rural) were the predictors. In the test set, the best model was the XGBoost model, which had an area under the curve of 0.942 [9]. It is crucial to note that these indicators are not routinely collected during the standard COPD diagnosis and treatment process, and there are no corresponding physiological indicators. This limitation diminishes the potential for their practical application. Moreover, the model does not provide specific tools that can be directly used in clinical settings. In order to address these key issues and fill this gap that is a frailty risk prediction model suitable for application by primary care physicians in the field of COPD. This study was based on the China Elderly Comorbidity Medical Database (CECMed) database and utilized interpretable machine learning methods to construct a novel risk prediction model for frailty in elderly patients with COPD. The aim was to provide relevant tools in order to optimize the management content during the stable phase of the disease.

Methods

Subjects

We established a bidirectional multicenter database of older adults with chronic diseases named CECMed, The proportion of the prospective cohort part is over 35%. The trial registration number (TRN) is NCT06316544 and study registration dates at 12 March 2024. The retrospective cohort was enrolled between January 2010 and January 2022, while the prospective cohort was enrolled between January 2023 and November 2023. Patients were recruited from selected tertiary hospitals and community hospitals in the southern, northern, and central regions of China, resulting in a database of 5866 patients. We selected patients with COPD based on specific criteria. Inclusion criteria included: (1) Meeting diagnostic criteria per the Global Initiative for Chronic Obstructive Lung Disease (GOLD) 2023 Report [10]; (2) being aged 65 years or older; (3) demonstrating adequate cognitive capacity to participated in; (4) Providing written informed consent. Exclusion criteria included: (1) Expected survival time of less than 3 months; (2) Presence of severe physical or cognitive impairments that affect communication or participation in the assessment; (3) Incomplete assessment using the FRAIL scale. The frailty assessment was conducted during the stable phase of the COPD.

This study adheres to the Declaration of Helsinki guidelines and is registered (www.clinicaltrials.gov, NCT06316544) and has been approved by the ethical review boards at the participating hospitals. All patients included in the study provided informed consent.

Methods

Data collection

The study’s demographic and clinical characteristics encompassed the following: (1) demographic and socioeconomic status such as age, gender, education level, living arrangements, smoking status and drinking status; (2) medical history including the Charlson Comorbidity Index (CCI), falls experienced in the preceding 12 months; (3) medication use (defined as polypharmacy when ≥ 5 medications were taken concurrently); (4)vital signs such as height, weight, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), heart rate, respiratory rate, pulse oxygen saturation (SpO2) and gait speed; (5) comprehensive geriatric assessment such as frailty, hearing, and vision. Data collection involved face-to-face interviews conducted by trained investigators. The investigators followed standardized procedures and scripts to ensure consistency in explaining the study’s objectives and data collection methods. Physiological measurements (e.g., blood pressure, BMI) adhered to WHO-STEP-wise protocols using calibrated instruments, while self-reported information was gathered through structured interviews. The measurement methods, recording procedures and entry standards for all of the above indicators were all determined as standardized collection items at the beginning of the database establishment, ensuring the consistency and comparability of data standards in both retrospective and prospective cohorts.

Frailty assessment

Frailty status was assessed using the FRAIL scale [11], a patient-reported tool developed by the International Association of Nutrition and Ageing (IANA) and presented in Table 1. Each item is scored 1 point, resulting in a total score ranging from 0 to 5, with individuals classified as frail (3–5 points), pre-frail (1–2 points), or non-frail (0 points). All the participants included in the analysis completed the FRAIL scale assessment; all the predictor variables in the study were independently collected and did not overlap with or conflict in sequence with the items of the FRAIL scale.

Table 1.

FRAIL scale

component	Question	Scoring
Fatigue	Are you fatigued? (Do you feel tired most or all of the time? )	No = 0, Yes = 1
Resistance	Do you have difficulty walking up one flight of stairs without assistance?	No = 0, Yes = 1
Aerobic	Do you have difficulty walking one block without assistance?	No = 0, Yes = 1
Illnesses	Do you have more than 5 illnesses? (Based on a list of 11 chronic diseases*)	No = 0, Yes = 1
Loss of weight	Have you lost more than 5% of your body weight in the past year? Scoring	No = 0, Yes = 1

*11 chronic diseases including hypertension, diabetes, cancer (other than minor skin cancer), chronic lung disease, heart attack, congestive heart failure, angina, asthma, arthritis, stroke, and kidney disease

Sample size calculation

The sample size for this cross-sectional study was calculated using the formula:

A meta-analysis of 36 studies showed that the prevalence of frailty in older patients with COPD was 36%(95%CI: 31%-40%) [12] (P = 0.36). When =1.96, =0.04, the initial estimation indicated a sample size of 554 participants. Factoring in potential attrition rates of 20–30% and accounting for missing data, the final recommended sample size ranges from 663 to 720 participants.

Statistical analysis

Data were analyzed using SPSS 26.0 and R 4.3.0. Continuous variables following a normal distribution were presented as mean ± SD and analyzed using Student’s independent samples t-test. Non-normally distributed continuous variables were reported as median (interquartile range) and analyzed with Mann-Whitney U tests. Categorical variables were presented as frequency (%) and analyzed with chi-square tests or Fisher’s exact tests. Variables with missing values exceeding 15% were excluded, and other missing data were imputed using the Multiple Imputation by Chained Equations (MICE) method with predictive mean matching (PMM). All statistical tests were two-tailed with α = 0.05.

The dataset was randomly divided into training and testing sets at a ratio of 7:3. Predictive variables were selected through Least Absolute Shrinkage and Selection Operator (LASSO) regression. The predictor factor selection is based on the combination of λ.min and λ.1se, and is determined after being evaluated by clinical experts, in order to balance statistical robustness and clinical practicability. Five machine learning (ML) algorithms were employed to construct predictive models: Logistic Regression (GLM), Random Forest (RF), Support Vector Machine (SVM), Recursive Partitioning (RP), and Neural Network (NNT). The best-performing model was selected based on the comparison of the area under the receiver operating characteristic (ROC) curve (AUC) values. Model performance was rigorously assessed in both the training and testing sets using calibration curve analysis, decision curve analysis (DCA). The optimal model was visually represented through a nomogram.

Results

Comparison of baseline characteristics

Following the application of predefined inclusion and exclusion criteria and the completion of data cleaning procedures, a total of 860 participants were included in the final analytical cohort, as illustrated in Fig. 1. All the subjects in the analysis cohort completed the FRAIL scale assessment. The cohort consisted of 562 males (65.3%) and 298 females (34.7%), with a mean age of 81.7 ± 9.34 years. Using FRAIL scale assessments, 308 individuals (35.8%) were classified as non-frailty (score of 0), 357 (41.5%) as pre-frailty (scores of 1–2), and 195 (22.7%) as frailty (score of ≥ 3). For comparative analysis, patients were divided into two groups: non-frailty (FRAIL score = 0, 35.8%) and pre-frailty/frailty (FRAIL score ≥ 1, 64.2%). A comparative analysis between these groups encompassed demographic characteristics, symptoms, physical signs, medication usage, and comorbidities, with detailed results presented in Table 2.

Fig. 1.

The workflow of the patient selection process

Table 2.

Baseline characteristics between frailty/Pre-frailty and non-frailty

	Pre-frailty and frailty (n = 552)	Non-frailty (n = 308)	/t/z	P
Age, years, mean SD	83.838.88	77.818.90	-9.514	<0.001
Gender, male, n(%)	355(64.3%)	207(67.2%)	0.732	0.392
Education level, n(%)			1.796	0.616
Primary school and lower	122(22.1%)	74(24.0%)
Junior high school	162(29.3%)	86(27.9%)
Senior high school	76(13.8%)	50(16.2%)
College and higher	192(34.8%)	98(31.8%)
Living alone, n(%)	123(22.3%)	37(4.3%)	13.768	<0.001
BMI, kg/m2, meanSD	23.923.60	24.463.54	2.146	0.032
Smoking status, n(%)	66(12.0%)	45(14.6%)	1.239	0.266
Drinking status, n(%)	91(16.5%)	65(21.1%)	2.840	0.092
Insomnia,, n(%)	163(29.5%)	43(14.0%)	26.303	<0.001
Falls in last year, n(%)	64(11.6%)	9(2.9%)	19.140	<0.001
Emergency or hospitalization in the last year, n(%)	145(26.3%)	61(19.8%)	4.533	0.033
CCI, median(IQR)	2(1, 3)	2(1, 2)	-5.616	<0.001
bronchodilator utilization, n(%)	176(31.9%)	120(39.0%)	4.386	0.036
Respiratory rate, times per minute, meanSD	20.072.49	19.572.26	-2.999	0.003
SBP, mmHg, median(IQR)	134(124, 145)	132(124, 142)	-1.894	0.058
DBP, mmHg, median(IQR)	70(65, 78)	73(68, 80)	-4.000	<0.001
Heart rate, times per minute, median(IQR)	76(69, 82)	75(68, 80)	-1.18	0.238
SpO2, %, median(IQR)	96(95, 98)	97(95, 98)	-2.092	0.036
Gait speed, meters per minute, meanSD	0.660.26	0.930.26	14.230	<0.001
Grip strength, kg, median(IQR)	20.5(20.5, 25)	26.5(21.5, 30.5)	-10.742	<0.001
Hearing impairment, n(%)	264(47.8%)	100(32.5%)	19.103	<0.001
Vision impairment, n(%)	339(61.4%)	175(56.8%)	1.736	0.188

Machine learning methods constructed the prediction model

After 10-fold cross-validation, the results of LASSO regression in the training set are shown in Fig. 2 (a and b), with Lambda.min = 0.0187 and Lambda.1se = 0.0571. The cross-validation error band is relatively narrow (Fig. 2b), indicating that the model selection results are stable and reliable. Based on the results of LASSO regression analysis, the predictive factors were selected by comprehensively considering statistical robustness and clinical practicability. Ultimately, six predictors were retained, encompassing age, CCI, bronchodilator utilization, DBP, gait speed, and grip strength. These six factors correspond to the six rightmost curves in Fig. 2a.

Fig. 2.

A: The coefficient profile plot from the log (lambda) sequence suggested 3 valid predictors before Lambda.1se and 13 valid predictors before Lambda.min; B: The partial likelihood deviance (binomial deviance) curve versus log (lambda) suggests that there are 6 or 7 valid predictors between Lambda.min and Lambda.1se

Ten rounds of 10-fold internal cross-validation were conducted to develop five ML models. The comparison of specific parameters of each model is presented in Table 3. Based on the AUC, the best model is Logistic Regression (GLM), with an AUC is 0.858 (95%CI: 0.815–0.897, 61.9% of sensitivity and 90.2% of specificity).The coefficients and P-values of each predictor in the logistic model were as follows: age (β = 0.0517, p < 0.001); CCI (β = 0.1257, p = 0.0711); bronchodilator utilization (β = -0.6094, p = 0.0047); DBP (β = -0.0614, p = 0.1353); gait speed (β = -3.2644, p < 0.001); and grip strength(β = -0.0845, p < 0.001). The ROC curve of each model is shown in Fig. 3a which the main difference in the middle and the best intercept point. Validation of GLM model stability across training and validation sets is shown in Fig. 3b evaluating the generalization of the optimal GLM model by comparing its ROC performance in the training set (red curve, AUC = 0.858, 95%CI: 0.815–0.897) and validation set (green curve, AUC = 0.851, 95%CI: 0.803–0.894). These two nearly overlapping curves and the extremely small difference in AUC (ΔAUC = 0.007) indicate the stability of the model during internal validation.

Table 3.

The comparison of specific parameters of each ML model

Model	Accuracy	AUC	Sensitivity	Specificity	Brier
GLM	0.810	0.858	0.619	0.902	0.146
NNT	0.779	0.851	0.690	0.822	0.149
RF	0.775	0.844	0.631	0.845	0.152
RP	0.752	0.757	0.643	0.805	0.176
SVM	0.202	0.856	0.381	0.115	0.522

Fig. 3.

A: ROC curves of five machine learning models: Logistic Regression (GLM, red), Random Forest (RF, blue), Support Vector Machine (SVM, orange), Regression Partition (RP, purple), and Neural Network (NNT, green). The horizontal axis represents 1-Specificity (false positive rate), and the vertical axis represents Sensitivity (true positive rate). The Area Under the Curve (AUC) values are annotated for each model and GLM demonstrated the highest classification accuracy; B: The ROC performance of optimal GLM model in the training set (red curve, AUC = 0.858) and validation set (green curve, AUC = 0.851)

Nomogram of GLM model

Taking the six independent predictors comprising age, CCI, bronchodilator utilization, DBP, gait speed, and grip strength to construct a nomogram by the GLM model (Fig. 4). As an example to better explain the nomogram model: if the subject is age of 80, CCI of 6, not using of bronchodilator, DBP of 50 mmHg, gait speed of 1 m/s, and Grip Strength of 19 kg. So the total points are 155, and the probability of pre-frailty or frailty is estimated similarly to be 86%.

Fig. 4.

Nomogram for predicting the risk of pre-frailty and frailty based on optimal GLM model: This nomogram integrates six clinical variables to estimate individualized risk scores. Each variable (Age, CCI, Bronchodilator use, DBP, Gait speed, and Grip Strength) is assigned a corresponding score on the “Points” axis (0-100) through vertical alignment. The total points (0-220, bottom axis) are calculated by summing individual variable scores and are converted to the predicted probability of pre-frailty/frailty (0.1–0.9, bottom probability axis)

GLM model performance evaluation

Following 2000 iterations of Bootstrap resampling in the training set, the resulting calibration curve exhibits strong alignment with the ideal reference line (Fig. 5a). High concordance between predicted and actual outcomes demonstrates robust calibration reliability. Decision curve analysis (DCA) (Fig. 5b) revealed a favorable net clinical benefit across a wide threshold probability range (0.25-1.0), underscoring its superior clinical utility.

Fig. 5.

A: The x-axis represents predicted probabilities (0–1.0), and the y-axis corresponds to observed probabilities. Ideal line (black dotted): Perfect calibration with predicted = observed probabilities. Apparent calibration (red solid line): Unadjusted model performance on training data, closely aligning with the ideal line. Bias-corrected curve (blue solid line): Bootstrap-adjusted calibration, maintaining proximity to the ideal line (< 2% deviation across probabilities), confirming minimal overfitting. B: DCA for clinical utility.The DCA evaluates net benefit (y-axis) against varying risk thresholds (x-axis, cost-benefit ratios from 1:100 to 100:1). Model(Train) (red): Net benefit for the training cohort. Model (Test) (green): Validation cohort consistency. All (gray): Base scenario of all intervention. None (gray): Base scenario of no intervention

Discussion

The prevalence of frailty among COPD patients increases with age and varies across studies due to differences in populations and frailty assessment tools [3]. A recent meta-analysis of 20 studies reported that the prevalence of frailty in COPD patients is 32.07% (95%CI:26.64%-37.49%), ranging from 6.43% to 71.70% depending on the specific frailty assessment tool utilized [6]. Our study found a frailty prevalence of 22.7%, aligning closely with the meta-analysis findings. Dias et al. employed the FRAIL scale in assessing frailty prevalence in 153 COPD patients, revealing rates of 50.3% (77/153) for frailty and 35.3% (54/153) for pre-frailty [13]. The higher prevalence reported by Dias may be attributed to the inclusion of more severe COPD cases in their study compared to our stable COPD patient cohort [13]. Similarly, Limpawattana reported that the prevalence rates of frailty and pre-frailty assessed by the FRAIL scale were 6.6% and 41.3% [14]. The prevalence rate observed in their study was lower than that in our study, potentially due to variations in patient age (≥ 18 years in their study versus ≥ 65 years in ours). In our previous study, we compared the performance of four frailty assessment tools, including Fried Frailty Phenotype (FFP), Clinical Frailty Scale (CFS), Frailty Index of Accumulative Deficits (FI-CD), and Short Physical Performance Battery (SPPB) in assessing frailty prevalence and predicting adverse outcomes in hospitalized older COPD patients. We observed different frailty prevalence in the study that FFP: 51.0%, CFS: 64.2%, FI: 59.6%, SPPB: 58.6% [15]. These results were similar to those of our study, indicating that different frailty assessment tools demonstrate comparable efficacy in evaluating frailty among COPD patients.

The primary risk factors for frailty in patients with COPD include predisposing factors such as genetic, environmental influences, diminished physiological reserves due to age and comorbidities, as well as other contributing factors like sedentary behavior, exercise limitations, smoking, polypharmacy, malnutrition, and abnormal weight status [5]. In addition to age and comorbidities, our research reveals that the use of bronchodilators, DBP, gait speed, and grip strength are significant determinants of frailty and pre-frailty in older COPD patients.

This study found that the use of bronchodilators is a protective factor against frailty in older COPD patients, suggesting that effective medication and management of COPD can lower the risk of frailty in this population. The European Respiratory Society emphasizes the importance of optimal COPD management in preventing frailty and in the care of individuals with post-frailty conditions. Even short-term pulmonary rehabilitation interventions have been shown to effectively reduce risk factors associated with frailty [5].

Our study identified a previously unreported association between low DBP and frailty/pre-frailty in older COPD patients.A large-scale study of 13,633 patients with acute exacerbation of COPD (AECOPD) found that low admission DBP (below 70 mmHg) was associated with poorer outcomes in hospitalized AECOPD patients, regardless of cardiovascular disease (HR = 2.16, 95% CI: 1.53–3.05, P < 0.01) [16]. Two mechanistic hypotheses may explain this association: Firstly, low DBP could result in myocardial hypoperfusion and injury while also reflecting end-organ perfusion status in eyes, kidneys, and brain [17]. Secondly, low DBP might signify the presence of advanced aortic atherosclerosis and a greater burden of cardiovascular disease, implying inferior baseline cardiovascular health [18]. However, additional validation is necessary to establish the causal link between low DBP and frailty in older COPD patients.

Gait speed and handgrip strength, two key components of the FFP, demonstrate significant predictive value for frailty and clinical outcomes in older COPD patients [19]. These two metrics elucidate the influence of COPD on frailty via three primary pathophysiological pathways: dysregulated muscle metabolism, chronic inflammation, and progressive decline in physical function [20]. Gait speed objectively reflects exercise capacity and gas exchange in COPD patients. Studies indicate that even a modest 0.1 m/s increase in gait speed is significantly associated with a 13.2% reduction in the risk of readmission (sHR=0.868, p=0.001) and a 25.3% decrease in mortality risk (sHR=0.747, p=0.002) among COPD patients [21]. The deterioration in grip strength represents a phenotypic manifestation of fundamental aging processes, encompassing sarcopenic muscle depletion, amplified cellular senescence pathways, and consequent functional limitations. These factors contribute to a detrimental cycle, leading to a decline in pulmonary function and a notable increase in frailty risk among older COPD patients[22]. Gait speed and grip strength are widely used functional assessment indicators in the field of geriatrics. Their measurement process is standardized, takes a short time (single test ≤ 2 min), and has low equipment costs. In clinical settings with basic training (such as community health centers, respiratory outpatient clinics), this measurement can be integrated into the routine follow-up process, with the additional time burden being manageable (approximately 3–5 min per patient).

Frailty in COPD patients is associated with a hastened progression of the disease, increased severity of symptoms, and a higher treatment burden. Early identification of frailty through predictive screening enables proactive interventions that can help maintain lung function, alleviate symptoms, and ultimately improve quality of life and clinical outcomes [23]. Evidence suggests that a composite threshold of the COPD Assessment Test (CAT) and Medical Research Council (MRC) dyspnea scale could predict frailty in COPD patients, with the AUC of 0.79 (95%CI: 0.72–0.86), 80.3% of specificity and 62.3% of sensitivity [13]. However, the intricate nature of these scales hampers their swift and straightforward application in routine clinical practice. A frailty prediction model was developed based on the 2018 CHARLS database included seven predictors: depression, smoking, gender, social activities, dyslipidemia, asthma, and residence type (urban vs. rural) [9]. The model was a XGBoost model, which had an AUC of 0.942 (95%CI: 0.925–0.959), an accuracy of 0.915 (87.3% of sensitivity, and 91.1% of specificity). However, the following aspects have limited its application. Firstly, depression and social activities require subjective assessment, while abnormal blood lipids need to be determined through laboratory tests. Secondly, the model construction uses XGBoost, which lacks interpretability, making it difficult for clinicians to understand the decision-making process of the model. Thirdly, the model cannot clearly establish the relationship between the standard treatment of COPD and weakness. Finally, the model only provides an online application tool, without offering specific tools that can be directly used in the clinical environment. In our research, we systematically gathered and validated frailty-related risk factors, constructing a nomogram model for predicting frailty and pre-frailty risk. Firstly, we utilized six clinical indicators commonly employed in the management of chronic obstructive pulmonary disease: age (demographic standard), CCI (clinical standard), bronchodilator usage (treatment standard), DBP (vital sign standard), gait speed and grip strength (functional assessment standard). These indicators are measured at each routine visit for chronic obstructive pulmonary disease in both primary and tertiary medical settings, enabling our model to be implemented immediately without increasing the clinical burden. Secondly, we employed logistic regression (GLM) and a clinically intuitive nomogram (Fig. 4), which shows the magnitude and direction of the impact of each predictor, and can be printed and placed in the patient’s medical record for rapid risk assessment during routine visits. Finally, the direct association between chronic obstructive pulmonary disease (COPD) management and frailty status. Our model clearly identified bronchodilator usage as a protective factor, and low DBP as a new risk factor, directly linking medication adherence and cardiovascular health to reduced frailty risk. This direct association enables clinicians to view frailty prediction as part of the continuum of chronic obstructive pulmonary disease management, rather than an isolated assessment. For example, patients with low diastolic blood pressure and poor bronchodilator usage may prioritize treatment options that combine cardiovascular optimization with chronic obstructive pulmonary disease management.

In clinical practice, this model is suitable for being applied in the routine follow-up of COPD patients during their stable periods, especially at key nodes such as the first diagnosis, annual assessment, or follow-up after the condition stabilizes. This model can be combined with the current chronic disease management programs for COPD in primary medical institutions to optimize the disease management content.

This study has several limitations. Firstly, although internal validation was performed using a 70:30 data split, external validation in larger, comparable cohorts is needed to confirm the model’s generalizability. Secondly, the applicability of this model may be limited due to the absence of laboratory biomarkers, pulmonary function data, and lung imaging data. Using the MICE method to supplement the missing data may affect the accuracy of the model. In subsequent iterations, these data can be incorporated to improve the accuracy of the model. Finally, the absence of long-term follow-up data hinders a comprehensive understanding of frailty’s impact on the short- and long-term outcomes of older Chinese COPD patients and the validation of the model’s effectiveness.

Conclusion

This study reveals that the combined prevalence of frailty and pre-frailty among older COPD patients is 64.19%, with 22.7% meeting frailty criteria. By integrating easily accessible clinical predictors such as age, CCI, bronchodilator utilization, DBP, gait speed, and grip strength, a logistic regression-based nomogram model was formulated, exhibiting strong predictive capability (AUC = 0.858, 95%CI: 0.815–0.897, 61.9% of sensitivity and 90.2% of specificity). For many relatively healthy COPD patients, early prediction of frailty can move the intervention window forward. The model’s simplicity and utilization of standard clinical parameters enhance its practicality for swift frailty screening in clinical settings. Notably, low DBP was identified as a new risk factor, potentially indicating cardiovascular compromise and reduced organ perfusion in this demographic. This research offers a practical tool for early frailty detection, aligning with global initiatives to enhance COPD management through proactive, comprehensive risk evaluation.

Authors' contributions

**Jie Li: ** Formal analysis, Methodology, Writing original draft, Writing review and editing; **Wen Tang: ** Funding acquisition, Methodology, Writing review and editing; **HuaYu Yang, Jia Luo, ** and **Yunli Xing: ** Formal analysis; **Ying Sun: ** Formal analysis, Conceptualization, Project administration, Supervision; **Qing Ma: ** Conceptualization, Formal analysis, Funding acquisition, Methodology, Project administration, Supervision, Writing review and editing.

Funding

This work was funded by National Science and Technology Major Project of the People’s Republic of China (2021ZD0111000).

Data availability

Restrictions apply to the availability of data generated or analyzed during this study to preserve participant confidentiality. The corresponding author will on request detail the restrictions and any conditions under which access to some data may be provided.

Declarations

Ethics approval and consent to participate

This study adheres to the Declaration of Helsinki guidelines and has been approved by the ethical review boards at Beijing Friendship Hospital, Capital Medical University(2022-P2-045-01). All patients included in the study provided informed consent.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.