Table of Contents
Study population
Our study recruited 251 male outpatients with stable COPD from October 2019 to December 2021 at the First Affiliated Hospital of Soochow University diagnosed according to the Global Initiative for Chronic Obstructive Lung Disease. Another 85 healthy participants during the same period were also included as the control group. They were randomly recruited by posting flyers on bulletin boards as well as from data collected at the hospital. Informed consent was obtained from all subjects and this study was approved by the research ethics committee of the first hospital affiliated to Soochow university (NO. 2022088).
Information collection
We collected the demographic and clinical characteristics of all participants, including their age, height, weight, history of smoking, acute exacerbations during the past year, and evaluation of symptoms with the COPD Assessment Test (CAT) and Modified British Medical Research Council (mMRC).
All the enrolled patients underwent spirometry using a Master screen-PFT system (Vyaire, Mettawa, Ill, USA) in the Pulmonary Function Test Room, and the forced expiratory volume during one second (FEV1) and forced vital capacity (FVC) were measured, according to the American Thoracic Society (ATS) consensus guidelines6. The FEV1, FEV1/FVC, and the ratio of FEV1 to predicted FEV1(FEV1%) after inhaling bronchodilators were recorded. According to the category of airflow limitation in COPD (based on FEV1 after the administration of a bronchodilator) in patients with FEV1/FVC < 0.70, stable COPD was divided into four GOLD grades: mild (GOLD 1), ≥ 80% of predicted FEV1; moderate (GOLD 2), 50–80% of predicted FEV1; severe (GOLD 3), 30–50% of predicted FEV1; very severe (GOLD 4), < 30% of predicted FEV1. 6-min walk test was performed according to guideline at the time of inclusion by investigator7. Meanwhile, the participants' handgrip strength was measured with a WCS-100 electronic handgrip dynamometer (produced by Shanghai Yilian Science and Education Equipment Co., LTD.), with the maximum value of the dominant hand recorded.
Body composition was assessed using a BCA-2A bioelectrical impedance body composition analyzer (Tsinghua Tongfang Co., Ltd., Beijing, China) by specialized technician in the mean time. Body weight, waist-to-hip ratio (WHR), fat-free mass (FFM), fat mass (FM), muscle mass, skeletal muscle mass (SMM), visceral fat area (VFA) and other body composite variables were measured. The FFMI was calculated as the FFM/(height)2. According to the European Society for Clinical Nutrition and Metabolism (ESPEN)8, individuals with a BMI below 20 kg/m2 if they are under 70 years of age, or below 22 kg/m2 if they are 70 years of age or older, or a FFMI below 17 kg/m2 in males, were classified as having malnutrition.
Variables selection and construction of nomogram
We developed the risk-prediction model for the occurrence of malnutrition in stable COPD patients following the transparent reporting of a multivariable-prediction model for Individual Prognosis or Diagnosis guidelines9. The variables included patients’ age, handgrip strength, 6-min walk distance (6MWD), WHR, acute exacerbations during the past year, the history of smoking, CAT scores, FVC and FEV1%. We log-transformed values of WHR that was not normally distributed for inclusion in the model. 6MWD was converted into categorical variables, and the roc curve was used to determine the optimal cut-off. Besides, we used a restricted cubic spline to flexibly model and verify the linear relationship between continuous variable and outcome. After transforming, the least absolute shrinkage and selection operator (LASSO) method was taken for variable selection in the training set. In the LASSO approach, the value of λ is used to penalize the absolute magnitude of a regression model's coefficients. The estimates approach zero more as the penalty is increased. As a result, irrelevant variables have zero coefficients, which can be used to eliminate non-influential predictors from the final model10.
The model was developed based on the selected features and refitted to avoid model overfitting. A nomogram was then constructed by using a linear combination of the selected features weighted by their regression coefficients. "Points" indicates the score of the corresponding factor below and "Total Points" indicates the summation of all the scores of factors above.
The discriminative ability of the nomogram was evaluated by using Harrell's concordance index (C-index), which was analogous to the area under the receiver operating characteristic curve (AUC). Further, a calibration curve was plotted and the Brier score was calculated to evaluate the agreement between the predicted and actual risk. When the Brier score ≤ 0.25, the model was considered to have favorable calibration. Internal validity and adjustment for overfitting of the nomogram were implemented with a bootstrap resampling (100 times) analysis.
Statistical analysis
Continuous variables were expressed as mean standard deviation or median interquartile range, while categorical values were expressed using relative frequencies and proportions. Comparisons of parameters between 2 different groups were conducted with the t-test and the Mann–Whitney U test for continuous variables with or without normal distribution. Categorical variables were evaluated using the chi-square test or Fischer exact test. Data were analyzed using SPSS software (Version 24.0), and R software (Version 4.1.0). A p-value < 0.05 was considered statistically significant.
