Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Development and validation of a prediction index for recent mortality in advanced COPD patients

Development and validation of a prediction index for recent mortality in advanced COPD patients www.nature.com/npjpcrm ARTICLE OPEN Development and validation of a prediction index for recent mortality in advanced COPD patients 1 2 3 3 4 3 3,5 Sheng-Han Tsai , Chia-Yin Shih , Chin-Wei Kuo , Xin-Min Liao , Peng-Chan Lin , Chian-Wei Chen , Tzuen-Ren Hsiue and 3,5 Chiung-Zuei Chen The primary barrier to initiating palliative care for advanced COPD patients is the unpredictable course of the disease. We enroll 752 COPD patients into the study and validate the prediction tools for 1-year mortality using the current guidelines for palliative care. We also develop a composite prediction index for 1-year mortality and validate it in another cohort of 342 patients. Using the current prognostic models for recent mortality in palliative care, the best area under the curve (AUC) for predicting mortality is 0.68. Using the Modified Medical Research Council dyspnea score and oxygen saturation to define the combined dyspnea and oxygenation (DO) index, we find that the AUC of the DO index is 0.84 for predicting mortality in the validated cohort. Predictions of 1-year mortality based on the current palliative care guideline for COPD patients are poor. The DO index exhibits better predictive ability than other models in the study. npj Primary Care Respiratory Medicine (2022) 32:2 ; https://doi.org/10.1038/s41533-021-00263-7 INTRODUCTION developed to predict long-term survival. They all lacked accuracy 20–23 when applied to short-term events of <12 months . Marin The prevalence of and mortality associated with chronic et al. validated a number of existing prognostic indices in a large obstructive pulmonary disease (COPD) have been increasing individual pooled data set (n = 3633) from multiple cohort studies annually . However, current treatments have been disappointing with different stages of COPD. These prognostic indices included in terms of controlling airflow obstructions and reducing 2–4 the original BODE, the modified BODE (replacing the 6-min walk mortality . Although palliative care is shown to be effective in distance (6MWD) with peak oxygen uptake V’ as % predicted), O2 patients with COPD, these patients have fewer opportunities to 5,6 the BODEx (replacing the 6MWD with exacerbations), the eBODE receive palliative care than patients with cancer . Jabbarian et al. (BODE plus exacerbations), the SAFE (SGRQ score, air-flow found that the failure to implement advance care planning (ACP) limitation, and exercise tolerance), the ADO, and the DOSE. All- in chronic diseases is mainly due to the complexity and cause mortality prediction at 12 months was assessed for these unpredictability of the disease , and the uncertainty of disease indices, where the indices determined to be optimal for prediction 8–12 trajectory is even greater in COPD than in cancer . In addition, was the ADO (C statistic = 0.70). Boeck et al. developed the B-AE- COPD patients typically want to know more about their prognosis D indices (BMI, acute exacerbations, dyspnea) for 2-year mortality 13,14 in the early stages . Therefore, enormous effort has been made in the PROMISE study, and external validation of the B-AE-D was to find indicators to predict a poor prognosis accurately. performed in COCOMICS and the COMIC study for 1-year all-cause Researchers have found many indicators related to various mortality (C statistic = 0.68 and 0.74, respectively). Therefore, none adverse outcomes for COPD, including patient age, body mass of these indices had the strong predictive ability for 1-year index (BMI), dyspnea, smoking status, exercise capacity, acute mortality. In addition, none of these models were developed with 15–17 exacerbation, symptoms, and biological indicators . Unfortu- the specific aim of predicting all-cause mortality in stable COPD nately, as was the case with the first proposed indicator, FEV , patients within 12 months. there was no optimal way to predict mortality based on the 25 To the best of our knowledge, Bloom et al. was the only 17,18 indicator . After the multisystem involvement characteristic of research group to develop indicators (the BARC index) for COPD became known, the focus was moved to composite predicting 1-year mortality with the aim of palliative care in 15,19 indicators to achieve better predictive outcomes . The earliest advanced COPD (C statistic = 0.78 and 0.70 for the development developed and most widely investigated multicomponent indica- and validation cohorts, respectively). The variables in the BARC tors included the Body-Mass Index, Airflow Obstruction, Dyspnea, only required routinely collected non-specialist information, and Exercise Capacity (BODE) Index , which was also recom- which, therefore, helped identify patients seen in primary care mended for predicting outcomes by the Global Initiative for institutions, but a total of 18 variables were required. Because no Chronic Obstructive Lung Disease (GOLD) . Later, numerous existing indices had strong enough predictive ability for 1-year different indices were developed, including the Dyspnea and mortality in clinical practice, and very few indices were developed Airflow Obstruction (ADO) Index, the Dyspnea, Obstruction, with the specific aim of predicting 1-year mortality for palliative Smoking, Exacerbation (DOSE) Index, and various modifications care in stable COPD. In this study, we aimed to validate the of the BODE index . However, most of these indices were currently recommended prediction indices for palliative care, we Division of General Medicine, Department of Internal Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. 2 3 Department of Public Health, College of Medicine, National Cheng Kung University, Tainan, Taiwan. Division of Pulmonary Medicine, Department of Internal Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. Department of Internal Medicine and Oncology, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. These authors contributed equally: Tzuen-Ren Hsiue, Chiung-Zuei Chen. email: chen96@mail.ncku.edu.tw Published in partnership with Primary Care Respiratory Society UK 1234567890():,; S.-H. Tsai et al. also developed a new predictive index for 1-year mortality in Predictive variables for palliative care hospitalized ambulatory COPD patients. In the first part of the study, we evaluated the predictive ability of the currently recommended variables for estimating 1-year mortality in the palliative care guideline for COPD. We selected METHODS several variables for building the predictive model based on a 30–32 Study design review of the currently recommended prediction variables . We conducted this cohort study in the National Cheng Kung The variables included (1) mMRC score = 4, (2) frequent, severe AE (two or more AEs requiring hospitalization in the preceding year), University Hospital (NCKUH) from August 2006 to December 2015. (3) hypoxemia (SpO < 90% in ambient air), (4) BMI < 21, and (5) The patients included in the present study were part of another predicted FEV < 30%. We used several combined indices to test previous study . The patients were eligible for inclusion if they the accuracy of the prediction for 1-year mortality. The patients had received regular management for COPD at our hospital for >1 were subdivided into four groups: Group 1 was defined as patients year prior to their recruitment. All patients were diagnosed with with frequent, severe AE in combination with severe dyspnea COPD by pulmonologists according to the GOLD guidelines for (mMRC = 4). Group 2 was defined as patients with frequent, diagnostic criteria . The criteria were as follows: age >40 years, severe AE in combination with SpO < 90% in ambient air. Group 3 typical symptoms, such as cough, dyspnea, wheezing, or chest was defined as patients with frequent, severe AE combined with tightness in combination with evidence of chronic airflow predicted FEV < 30%. Group 4 was defined as patients with obstruction, as defined by a postbronchodilator ratio of forced frequent, severe AE in combination with BMI < 21. expiratory volume in 1 s (FEV ) to a forced vital capacity (FVC) of <70%. Pulmonary function tests were performed following the Modeling the predictive scores standard protocols of the American Thoracic Society . All patients Because of the generally unsatisfactory predictive power found in were enrolled under clinically stable conditions. We excluded previous studies and with validating our results, we wanted to patients who were unwilling to participate and those who had derive a new predictive model for 1-year mortality from the advanced lung cancer and pulmonary fibrosis because of patient variables, including age, sex, BMI, disease severity, such as anticipated death in the near future. Patients with missing data mMRC dyspnea score, FEV , SpO , and comorbidities. The 1 2 and those lost to follow-up in the first year were also excluded variables were evaluated using multivariate Cox regression models from the analysis. In total, 752 patients with COPD were analyzed with a forward entering approach and a 5% significance level for (Supplementary Fig. 1). The Institutional Review Board of NCKUH the selection criteria. Significant regression coefficients were approved this study before commencement (IRB number: B-ER- converted to exponential expressions for the weighting of the 105-386 and B-ER-98-289). Written informed consent was variables used for the predictive indices. obtained for all participants while enrollment. Validation of the predicting index Prognostic variables and outcome To validate the predictive performance of our model, we selected A total of 752 consecutive COPD patients were recruited. All a second cohort. All patients in the development group were patients were monitored through December 2016 or until death. recruited from pulmonary outpatient departments. Considering We acquired age, smoking history, BMI, the severity of dyspnea that if the validation group and the developmental group assessed by grade on the modified Medical Research Council exhibited high homogeneity, it was expected that the proposed model would obtain very similar results for the two groups of (mMRC) dyspnea scale , the degree of comorbidity as evaluated patients. Patients in the validation group were recruited by using the Charlson index , oxygen saturation levels as detected screening individuals who had been diagnosed with COPD, not by pulse oximetry in room air (SpO ), and status of long-term only in the pulmonary outpatient department but also in the home oxygen usage from every patient at the time of inclusion as Center for Hospice Palliative Shared Care at NCKUH from July 2012 determined by research assistants in the study. Comorbidity was to August 2019. All patients were aged ≥40 years; COPD was evaluated using the Charlson index and included congestive heart defined according to the GOLD diagnostic guidelines and criteria failure, coronary artery disease, systemic hypertension, peptic as the developmental group; patients with advanced lung cancer ulcer, and diabetes mellitus as identified from the patient files and or pulmonary fibrosis were excluded. The date of recruitment of detailed interviews. A severe acute exacerbation of COPD was some patients from the Center for Hospice Palliative Shared Care defined as an acute event characterized by a worsening of the overlapped with the time periods during which the development patient’s respiratory symptoms that were beyond day-to-day group was recruited. These patients were not excluded from this variations that also required hospitalization. The number of severe study since the source of patients was different from that for the exacerbations in the preceding year was recorded by research development group (Center for Hospice Palliative Shared Care assistants according to the patient’s chart as the primary means of versus the pulmonary outpatient department). All patients had data collection; self-reported data was used to supplement complete follow-up for 1 year or until death. this data. All-cause mortality was defined as the endpoint of the study. Statistical analysis The survival status of all patients was evaluated using a Continuous variables are presented as the median and inter- prospective observation, as reported in a previous study . All quartile range because the number of deaths was not large and patients were contacted during regular clinic visits or by therefore may not follow a normal distribution. Therefore, telephone interviews (if they missed an appointment). Most comparisons between survivors and nonsurvivors were performed patients who died during the study period had been regularly using Mann–Whitney U-test. Comparisons between categorical followed and had visited the hospital for treatment before their variables were performed using chi-square tests or Fisher’s death. Their dates of death were recorded and verified using exact tests. hospital records. Research assistants obtained the date of death of Kaplan–Meier survival curves and log-rank tests were used for patients who died outside the hospital by telephone contact with comparing different predictive variables. The ability to predict partners or family members. Survival status was also verified mortality within 1 year was analyzed using logistic regression through linkage with the Taiwan National Mortality Registry. models and the receiver operating characteristic (ROC) curve to npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK 1234567890():,; S.-H. Tsai et al. calculate the area under the curve (AUC). Data processing and severe AE and predicted FEV < 30% was 0.60, and the AUC for analyses were performed using the SPSS for Windows version patients with frequent, severe AE and BMI < 21 was 0.68. The ROC 17.0 statistical software (IBM, Armonk, NY, USA). for the different composite indices did not differ significantly between the four groups of combinations (Fig. 1). Kaplan–Meier survival analysis for all-cause mortality showed Reporting summary the 1-year survival rates for groups 1–4 were 62, 75, 88, and 78%, Further information on research design is available in the Nature respectively (Fig. 2). All indices, including those for groups 1–4, Research Reporting Summary linked to this article. showed high specificity but unsatisfactory sensitivity. The composite indices for group 1 with frequent, severe AE and RESULTS severe dyspnea had a better positive predictive value than the Participants other groups (Table 2). We enrolled 752 COPD patients from August 2006 through December 2015. The mean age of the patients was 70.6 years, and Development and validation of a prediction index most of them were men (92.7%). Twenty-eight percent of the In the univariate analysis, there were significant differences patients had severe to very severe airflow limitations (FEV < 50% between survivors and nonsurvivors in age, FEV , SpO , mMRC, of predicted); 25.2% of the patients had mMRC scores from 3 to 4, 1 2 BMI, anemia, and concurrent malignancies. Using multivariate and 50.7% had at least one severe AE in the preceding year during enrollment. At the end of the follow-up period in December 2016, regression, only mMRC and SpO were independent risk factors 378 patients had died (50.3%), and 60 patients (8%) had died for predicting 1-year mortality. within 1 year after the start of follow-up. We refined the dyspnea and oxygenation (DO) index by The baseline characteristics of survivors and nonsurvivors are weighting dyspnea and SpO based on the results of the shown in Table 1. Compared with survivors, nonsurvivors were multivariate regression model. We used the integers closest to older, had a worse pulmonary function, lower BMI, lower oxygen the hazard ratio for scoring predictive variables (Table 3). saturation, worse symptoms of dyspnea (higher mMRC score), and Compared with groups 1–4, the DO index had better discrimina- more AEs in the previous year. tion for mortality (AUC = 0.73; Fig. 1). We also performed a sensitivity analysis for patients with severe or very severe Predictive ability of the currently recommended models obstruction (FEV < 50%), where the DO index performed better The AUC values for predicting 1-year mortality in patients with (AUC = 0.81). The survival rates for the different DO scores are severe dyspnea (mMRC = 4) and patients having frequent, severe AE were 0.62 and 0.60, respectively. Combining predictor variables for patients with severe dyspnea and frequent severe AE was better than using only one variable (AUC = 0.68). The AUC for predicting 1-year mortality for patients with frequent, severe AE and SpO < 90% was 0.66. The AUC for patients with frequent, Table 1. Demographic and patient characteristics of survivors and nonsurvivors. Characteristic Survivors Nonsurvivors p Value (n = 692) (n = 60) Age, median (IQR) 71.2 (64.6, 78.7) 78.4 (72.5, 81.6) <0.01 Male n (%) 640 (92.5) 57 (95.0) 0.61 Fig. 1 ROC curve for severe dyspnea, severe acute exacerbation, Current smoker, 189 (27.3) 13 (21.7) 0.30 and different combinations of predictors and DO index for 1-year n (%) mortality in COPD patients. The AUC values for patients with severe Smoking quantity 45 (23, 70) 50 (20, 62) 0.92 dyspnea (mMRC = 4), frequent severe AE, groups 1–4, and DO index (pack-years) were 0.62, 0.60, 0.68, 0.66, 0.60, 0.68, and 0.73, respectively. FEV % 64 (48, 82) 50 (34, 64) <0.01 BMI 23.3 (20.5, 25.8) 20.5 (17.0, 24.5) <0.01 SpO % 97.0 (95.0, 98.0) 95.5 (92.0, 97.0) <0.01 CI score 2.0 (1.0, 3.0) 3.0 (1.0, 5.0) <0.01 Severe AE ≥ 2, n (%) 111 (16.0) 22 (36.7) <0.01 6MWT (meter) 344.0 (248.0, 400.0) 278.0 (206.0, 313.0) 0.11 SGRQ score 33.22 (18.2, 51.0) 59.13 (46.4, 65.3) <0.01 mMRC= 4 26 (3.7) 17 (28.3) <0.01 LTOT, n (%) 71 (10.3) 15 (25.0) <0.01 Discrete data are presented as number (percentage), and continuous variables are presented as median (IRQ). FEV forced expiratory volume in 1 s, BMI body mass index, SpO oxygen 1 2 saturation (%) detected with pulse oximeter when breathing in room air, CI Charlson index, severe AE ≥ 2 history more than one acute exacerbation that Fig. 2 Kaplan–Meier survival curves for 1-year mortality accord- required hospitalization in the preceding year, 6MWT 6 min walking test, ing to different recommended prediction indices for palliative SGRQ St. George’s Respiratory Questionnaire, mMRC modified Medical care. The 1-year survival rates were 62, 75, 88, and 78% for groups Research Council Dyspnea Scale, LTOT long-term oxygen therapy. 1–4, respectively. Published in partnership with Primary Care Respiratory Society UK npj Primary Care Respiratory Medicine (2022) 2 S.-H. Tsai et al. Table 2. Predictive accuracy of different recommended palliative care indices for 1-year mortality. Prognostic index Sensitivity Specificity PPV NPV Accuracy AUC mMRC= 4 28.3% 96.2% 39.5% 93.9% 90.8% 0.623 Severe AE ≥ 2 36.7% 84.0% 16.5% 93.9% 80.2% 0.603 Group 1 13.3% 98.1% 38.1% 92.9% 91.4% 0.684 Group 2 3.3% 99.1% 25.0% 92.2% 91.5% 0.657 Group 3 3.3% 98.0% 12.5% 92.1% 90.4% 0.634 Group 4 18.3% 94.4% 22.0% 93.0% 88.9% 0.679 mMRC modified Medical Research Council Dyspnea Scale in stable condition, Severe AE ≥ 2 more than one acute exacerbation that required hospitalization in the preceding year, Group 1 mMRC = 4 + severe AE ≥ 2, Group 2 severe AE ≥ 2 + SpO < 90%, Group 3 severe AE ≥ 2 + FEV < 30%, Group 4 severe AE ≥ 2 + BMI 2 1 < 21, PPV positive predictive value, NPV negative predictive value, AUC area under the curve, SpO oxygen saturation (%) detected with a pulse oximeter when breathing room air, FEV forced expiratory volume in 1 s, BMI body mass index. Table 3. Weighting of variables in DO index. Table 4. Survival analysis of 1-year mortality for different DO index scores of patients with severe and very severe COPD (n = 180). Variable β Adjusted HR Score Score Survived (n) Died (n) Survival Statistics p Value SpO (%) 2 rate (%) (chi-square) 95–100 0 1 1 DO 58.61 <0.001 90–94 0.55 1.7 2 2 56 2 96.55 85–89 1.05 2.9 3 3 58 2 96.67 <85 2.00 7.3 7 4 30 3 90.91 mMRC score 5 2 0 100.00 0–20 1 1 8 1 0 100.00 3 0.64 1.9 2 90 1 0 4 2.21 9.1 9 10 10 2 83.3 Coding according to the regression coefficient for DO index construction. 11 4 4 50 DO dyspnea and oxygenation. 12 1 2 33.3 16 0 2 0 shown in Table 4. In the group with the most severe COPD Score Survived (n) Died (n) Survival Statistics p Value (DO score = 12–16), the survival rate was only 20%. rate (%) (chi-square) We enrolled a total of 342 patients for the validation group. The patients in the validation group were older (73.5 vs. 72.2%), had DO 52.98 <0.001 lower oxygenation (96 vs. 97%), more comorbidities, as evaluated 2–7 147 7 95.45 by the CI score (4.0 vs. 2.0%), more symptoms, as evaluated by the 8–11 14 7 66.7 percentage of patients with mMRC = 4 (14.3 vs. 6.3%), and higher 1-year mortality (14.3 vs. 8.0%), than the developmental group 12–16 1 4 20.00 (Supplementary Table 1). When applying our DO score for Wilcoxon test. predicting 1-year mortality, the AUC was 0.84. In the group with b Log-rank test. the most severe COPD (DO score = 12–16), the positive and negative predictive values were 87% and 89%, respectively, for 1-year mortality. guideline-recommended prediction tool. There was no process for sample size calculation in the study. Therefore, we calculated the statistical power backward using our sample size. According to DISCUSSION 23–25 the predictive ability of existing indices in a previous review , This study showed that a combination of mMRC and frequent, we considered an AUC of 0.65 as the median discrimination power severe AE as a predictor of 1-year mortality demonstrated similar for previous predictors. Using our population of a total of 342 poor discrimination power as other combinations of factors. These patients in the validation group for a two-sided z-test at a factors included predictors with desaturation, or a poor grade of significance level of 0.05, we achieved a 99% power to detect the lung function, or low BMI combined with frequent, severe AE. The AUC between the median discrimination power of previous AUC values of these combinations ranged from 0.60 to 0.68. In predictors and the DO index in this study (PASS Power Analysis addition, all combined groups exhibited lower sensitivity but and Sample Size Software, NCSS, LLC., Kaysville, Utah). However, higher specificity. Therefore, these indices did a good job of ruling there were only 18 deaths out of 180 patients when the DO index outpatients who would survive for >1 year but tended to miss was applied in severe COPD patients (see Table 4). The small patients who would die within 1 year. numbers in some lattices may have thus affected the accuracy of Using the mMRC dyspnea score and oxygen saturation (SpO ), we developed a better discriminating model for 1-year mortality, the estimate. For example, when using a DO score = 9 as a cutoff value, the 1-year survival rate for patients with DO scores ≥ 9 was the DO index. The AUC value of the DO index was 0.73 for the prediction of 1-year mortality, and the AUC was 0.84 in the estimated to be 58% (15/26). The accuracy of the predictive ability, validated cohort, which was superior to the current palliative sensitivity, specificity, positive predictive value, and negative npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK S.-H. Tsai et al. predictive value was 87.8% (95% CI (confidence interval), and spirometry results, and the deaths were verified by linking 82.1–92.2%), 61.1% (95% CI, 35.8–82.7), 90.7% (95% CI, with a database from the Taiwan National Mortality Registry. In 85.2–94.7), 42.3% (95% CI, 28.6–57.4), and 95.5% (95% CI, addition, we performed a sensitivity analysis by excluding a small 92.2–97.4), respectively. The low precision for sensitivity and the amount of missing mMRC values and by restricting the subjects to positive predictive value as indicated by wider confidence patients with severe airflow limitations. Both results showed intervals was attributed to the smaller sample size in this test. similar findings, which indicated the consistency of the discrimi- In general, the discriminative power of the model for the native validity. Finally, we validated the DO index and showed its development group was better than that for the validation group. good predictive ability for 1-year mortality in the second cohort of Patients in the validation group were older, had lower oxygena- 342 patients. tion, more comorbidities, more symptoms, and higher mortality There are limitations to our study. First, we did not have a large than the development group in this study. The heterogeneity patient population and were only limited to one medical center. between our developmental and validation groups was one of the The second limitation of the present study was that few women reasons explaining why the predictive ability was better in the (7.3%) were enrolled. Recently, a large, real-world, cohort study validation group than in the development group. To evaluate revealed gender differences among COPD patients, where COPD whether the predictive ability of the DO index becomes stronger was more frequent among women (53.8%), and the overall over time, we analyzed the predictive ability for 3- and 5-year mortality rate was higher in men as compared with women (45 vs. mortality and found that the AUC values were 0.66 and 0.67, 38%). However, no differences in mortality due to COPD related to respectively. These results implied that the DO index is not gender were found . The majority of participants with smoking- suitable for predicting 3- and 5-year mortality. related COPD in Taiwan are male ; the smoking prevalence for The findings of poor discrimination for the current models in women is <5% in Taiwan. In contrast to industrialized countries in this study were consistent with a previous systemic review the West, COPD morbidity remains male predominant in most 33 37–39 conducted by Almagro et al. . They used indicators already Asian countries . However, the result of this study cannot be developed in previous articles to validate the performance in their directly generalizable to other countries due to this limitation. cohort. Composite indices were better than a single parameter, The third limitation was that socioeconomic status (SES) was not and the best AUC was 0.68 for the CODEX index (comorbidity, included in our predictive model. SES disadvantages appear to obstruction, dyspnea, and previous exacerbation). The author have a significant impact on COPD mortality and morbidity, where concluded that no single index is good enough to guide the individuals with the lowest SES consistently had been shown to initiation of palliative care. Thus, the clinician should not make this have significantly higher mortality than those with the highest 40,41 decision based solely on a predictive tool. However, the use of the SES . We thus suggest including an SES measurement in the proposed DO index improved the predictive power (AUC = 0.73) predictive model in further studies. for 1-year mortality, and the AUC was 0.84 in the validated cohort. In conclusion, this study demonstrated that the predictive The BARC index for prognostic factors, including BMI and blood values for 1-year mortality were poor, based on the current results (B), age (A), respiratory variables (airflow obstruction, recommendations for palliative care among COPD patients, exacerbations, smoking) (R), and comorbidities (C) was conducted including four different composite indices. The newly developed based on medical databases and had a satisfactory AUC for 1-year DO index proposed in this work exhibited better predictive ability mortality (AUC = 0.79). However, it included 18 variables in the than other alternatives. We suggest that COPD patients with DO model, such as age, BMI, FEV , severe exacerbations, smoking index scores ≥12, for example, patients with mMRC = 4 and SpO 1 2 status, multiple comorbidities, hemoglobin, platelets, and others < 90%, are good candidates to receive palliative care. for the evaluation . In contrast, the DO index proposed in this study is simple to use because only two clinical parameters, DATA AVAILABILITY dyspnea score and oxygenation, detected with a pulse oximeter, Full data sets are not available publicly currently for protecting patient privacy. But are needed. Another composite index, the ProPal-COPD tool, had the data can be requested reasonably to the corresponding author. a good predictive ability for 1-year mortality with an AUC of 0.82. This model relied on the following seven predictors: (1) a surprise question, (2) MRC dyspnea, (3) the Clinical COPD Questionnaire Received: 6 November 2020; Accepted: 25 November 2021; (CCQ), (4) FEV % of predicted value, (5) BMI, (6) previous hospitalizations for AECOPD, and (7) specific comorbidities. However, some variables, such as the surprise question and CCQ, were not always routinely captured. REFERENCES The DO index developed in this study was composed of the 1. Rabe, K. F. et al. Global strategy for the diagnosis, management, and prevention mMRC score and oxygen saturation. In a recent systematic review of chronic obstructive pulmonary disease: GOLD executive summary. Am. J. of predictive indicators in COPD, 24 models used composite Respir. Crit. Care Med. 176, 532–555 (2007). indicators . The dyspnea score (mMRC) is one of the ten most 2. Tashkin, D. P. et al. A 4-year trial of tiotropium in chronic obstructive pulmonary used parameters. Nishimura et al. also demonstrated that the disease. N. Engl. J. Med. 359, 1543–1554 (2008). severity of dyspnea was a more favorable predictor of death than 3. Yang, I. A., Fong, K. M., Sim, E. H., Black, P. N. & Lasserson, T. J. Inhaled corti- FEV . This result was consistent with the use of mMRC in this costeroids for stable chronic obstructive pulmonary disease. Cochrane Database article. Our model also included another indicator, SpO , which Syst. Rev. CD002991 (2007). was not commonly used in previous studies. Instead, some studies 4. Calverley, P. M. et al. Salmeterol and fluticasone propionate and survival in used arterial oxygen partial pressure (PaO ) as a predictor. PaO is chronic obstructive pulmonary disease. N. Engl. J. Med. 356, 775–789 (2007). 2 2 5. Au, D. H., Udris, E. M., Fihn, S. D., McDonell, M. B. & Curtis, J. R. Differences in most often used during hospitalization, where PaO can fluctuate health care utilization at the end of life among patients with chronic obstructive due to many factors, such as oxygen use, pneumonia, or pulmonary disease and patients with lung cancer. Arch. Intern. Med. 166, 326–331 cardiovascular instability. Additionally, drawing arterial blood (2006). may also lead to some local complications. In our study, we used 6. Jabbarian, L. J. et al. Advance care planning for patients with chronic respiratory SpO in stable patients breathing ambient air to measure constant diseases: a systematic review of preferences and practices. Thorax 73, 222–230 oxygenation status. Lower levels of invasiveness are also preferred (2018). in outpatient settings. 7. Higginson, I. J. et al. An integrated palliative and respiratory care service for The strength of this study is its diagnostic and measurement patients with advanced disease and refractory breathlessness: a randomized accuracy. COPD was diagnosed according to standard evaluations controlled trial. Lancet Respir. Med. 2, 979–987 (2014). Published in partnership with Primary Care Respiratory Society UK npj Primary Care Respiratory Medicine (2022) 2 S.-H. Tsai et al. 8. Crawford, A. Respiratory practitioners’ experience of end-of-life discussions in 35. Bellou, V., Belbasis, L., Konstantinidis, A. K., Tzoulaki, I. & Evangelou, E. Prognostic COPD. Br. J. Nurs. 19, 1164–1169 (2010). models for outcome prediction in patients with chronic obstructive pulmonary 9. Gott, M. et al. Barriers to advance care planning in chronic obstructive pulmonary disease: systematic review and critical appraisal. BMJ https://doi.org/10.1136/bmj. disease. Palliat. Med. 23, 642–648 (2009). l5358 (2019). 10. Scheerens, C. et al. “A palliative end-stage COPD patient does not exist”: a qua- 36. Lisspers, K. et al. Gender differences among Swedish COPD patients: results from litative study of barriers to and facilitators for early integration of palliative home the ARCTIC, a real-world retrospective cohort study. NPJ Prim. Care Respir. Med. care for end-stage COPD. NPJ Prim. Care Respir. Med. https://doi.org/10.1038/ https://doi.org/10.1038/s41533-019-0157-3 (2019) s41533-018-0091-9 (2018). 37. Jamrozik, E. & Musk, A. W. Respiratory health issues in the Asia-Pacific region: an 11. Gaspar, C., Alfarroba, S., Telo, L., Gomes, C. & Barbara, C. End-of-life care in COPD: overview. Respirology 16,3–12 (2011). a survey carried out with Portuguese pulmonologists. Rev. Port. Pneumol. 20, 38. Tan, W. C. & Ng, T. P. COPD in Asia: where East meets West. Chest 133, 517–527 123–130 (2014). (2008). 12. Smith, T. A. et al. Specialist respiratory physicians’ attitudes to and practice of 39. Tan, W. C. Trends in chronic obstructive pulmonary disease in the Asia-Pacific advance care planning in COPD. A pilot study. Respir. Med. 108, 935–939 regions. Curr. Opin. Pulm. Med. 17,56–61 (2011). (2014). 40. Gershon, A. S., Dolmage, T. E., Stephenson, A. & Jackson, B. Chronic obstructive 13. MacPherson, A., Walshe, C., O’Donnell, V. & Vyas, A. The views of patients with pulmonary disease and socioeconomic status: a systematic review. COPD 9, severe chronic obstructive pulmonary disease on advance care planning: a 216–226 (2012). qualitative study. Palliat. Med. 27, 265–272 (2013). 41. Sahni, S., Talwar, A., Khanijo, S. & Talwar, A. Socioeconomic status and its rela- 14. Janssen, D. J. A., Spruit, M. A., Schols, J. M. G. A. & Wouters, E. F. M. A call for high- tionship to chronic respiratory disease. Adv. Respir. Med. 85,97–108 (2017). quality advance care planning in outpatients with severe COPD or chronic heart failure. Chest 139, 1081–1088 (2011). 15. Dijk, W. D. et al. Multidimensional prognostic indices for use in COPD patient ACKNOWLEDGEMENTS care. A systematic review. Respir. Res. https://doi.org/10.1186/1465-9921-12-151 The study was funded by grants from the Ministry of Science and Technology (2011). (MOST107-2627-M-006-007, MOST 109-2314-B-006-091, MOST 110-2314-B-006-099), 16. Man, S. F. et al. C‐reactive protein and mortality in mild to moderate chronic and it was supported in part by the Higher Education Sprout Project, Ministry of obstructive pulmonary disease. Thorax 61, 849–853 (2006). Education to the Headquarters of University Advancement at National Cheng Kung 17. Nishimura, K., Izumi, T., Tsukino, M. & Oga, T. Dyspnea is a better predictor of University (NCKU). We are grateful to the Health and Welfare Data Science Center of 5-year survival than airway obstruction in patients with COPD. Chest 121, the Ministry of Health and Welfare and the Center of hospice palliative shared care at 1434–1440 (2002). National Cheng Kung University Hospital (NCKUH) for providing all the data sets, 18. Casanova, C. et al. Differential effect of modified medical research council dys- facilities, and linkage services that were required for this study. Parts of our results pnea, COPD assessment test, and clinical COPD questionnaire for symptoms were submitted as an e-poster in the ATS 2020 Virtual content. evaluation within the new GOLD staging and mortality in COPD. Chest 148, 159–168 (2015). 19. Celli, B. R. Predictors of mortality in COPD. Respir. Med. 104, 773–779 (2010). AUTHOR CONTRIBUTIONS 20. Celli, B. R. et al. The body-mass index, airflow obstruction, dyspnea, and exercise T.-R.H. and C.-Z.C. were involved in the concept development and study design. All capacity index in chronic obstructive pulmonary disease. N. Engl. J. Med. 350, authors contributed to the data retrieval and patient recruitment. S.-H.T. and C.-Y.S. 1005–1012 (2004). conducted the analysis. Paper drafting was mainly accomplished by S.-H.T. and C.-Z.C. 21. Vestbo, J. et al. Global strategy for the diagnosis, management, and prevention of and all authors contributed to the revision and approval of the final manuscript. chronic obstructive pulmonary disease: GOLD executive summary. Am. J. Respir. Crit. Care Med. 187, 347–365 (2013). 22. Smith, L. E. et al. Prognostic variables and scores identifying the end of life in COMPETING INTERESTS COPD: a systematic review. Int. J. Chron. Obstruct. Pulmon. Dis. 12, 2239–2256 (2013). The authors declare no competing interests. 23. Marin, J. M. et al. Multicomponent indices to predict survival in COPD: the COCOMICS study. Eur. Respir. J. 42, 323–332 (2013). 24. Boeck, L. et al. Prognostic assessment in COPD without lung function: the B-AE-D ADDITIONAL INFORMATION indices. Eur. Respir. J. 47, 1635–1644 (2016). Supplementary information The online version contains supplementary material 25. Bloom, C. I., Ricciardi, F., Smeeth, L., Stone, P. & Quint, J. K. Predicting COPD 1-year available at https://doi.org/10.1038/s41533-021-00263-7. mortality using prognostic predictors routinely measured in primary care. BMC Med. https://doi.org/10.1186/s12916-019-1310-0 (2019). Correspondence and requests for materials should be addressed to Chiung-Zuei 26. Chen, C. Z. et al. Using post-bronchodilator FEV1 is better than pre- Chen. bronchodilator FEV1 in evaluation of COPD severity. COPD 9, 276–280 (2012). 27. Blonshine, S., Mottram, C. D. & Wanger, J. Pulmonary Function Laboratory Man- Reprints and permission information is available at http://www.nature.com/ agement and Procedure Manual 2nd edn (American Thoracic Society, 2005). reprints 28. Bestall, J. C. et al. Usefulness of the medical research council (MRC) dyspnea scale as a measure of disability in patients with chronic obstructive pulmonary disease. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims Thorax 54, 581–586 (1999). in published maps and institutional affiliations. 29. Charlson, M. E., Pompei, P., Ales, K. L. & MacKenzie, C. R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J. Chronic Dis. 40, 373–383 (1987). 30. Lanken, P. N. et al. ATS End-of-Life Care Task Force. An official American Thoracic Open Access This article is licensed under a Creative Commons Society clinical policy statement: palliative care for patients with respiratory Attribution 4.0 International License, which permits use, sharing, diseases and critical illnesses. Am. J. Respir. Crit. Care Med. 177, 912–927 (2008). adaptation, distribution and reproduction in any medium or format, as long as you give 31. O'Donnell, D. E. et al. Canadian Thoracic Society recommendations for man- appropriate credit to the original author(s) and the source, provide a link to the Creative agement of chronic obstructive pulmonary disease - 2008 update. Can. Respir. J. Commons license, and indicate if changes were made. The images or other third party 15,1A–8A (2008). material in this article are included in the article’s Creative Commons license, unless 32. Thomas, K., Wilson, J. A. & GSF Team. GSF PIG 6th Edition. National Gold Stan- indicated otherwise in a credit line to the material. If material is not included in the dards Framework Centre in end of life care. http://www.goldstandardsframework. article’s Creative Commons license and your intended use is not permitted by statutory org.uk (2016) regulation or exceeds the permitted use, you will need to obtain permission directly 33. Almagro, P. et al. Palliative care and prognosis in COPD: a systematic review with from the copyright holder. To view a copy of this license, visit http://creativecommons. a validation cohort. Int. J. Chron. Obstruct. Pulmon. Dis. 12, 1721–1729 (2017). org/licenses/by/4.0/. 34. Yohannes, A. M., Baldwin, R. C. & Connolly, M. J. Predictors of 1-year mortality in patients discharged from hospital following acute exacerbation of chronic obstructive pulmonary disease. Age Ageing 34, 491–496 (2005). © The Author(s) 2022 npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png npj Primary Care Respiratory Medicine Springer Journals

Development and validation of a prediction index for recent mortality in advanced COPD patients

Loading next page...
 
/lp/springer-journals/development-and-validation-of-a-prediction-index-for-recent-mortality-bty80iJhFI
Publisher
Springer Journals
Copyright
Copyright © The Author(s) 2022
eISSN
2055-1010
DOI
10.1038/s41533-021-00263-7
Publisher site
See Article on Publisher Site

Abstract

www.nature.com/npjpcrm ARTICLE OPEN Development and validation of a prediction index for recent mortality in advanced COPD patients 1 2 3 3 4 3 3,5 Sheng-Han Tsai , Chia-Yin Shih , Chin-Wei Kuo , Xin-Min Liao , Peng-Chan Lin , Chian-Wei Chen , Tzuen-Ren Hsiue and 3,5 Chiung-Zuei Chen The primary barrier to initiating palliative care for advanced COPD patients is the unpredictable course of the disease. We enroll 752 COPD patients into the study and validate the prediction tools for 1-year mortality using the current guidelines for palliative care. We also develop a composite prediction index for 1-year mortality and validate it in another cohort of 342 patients. Using the current prognostic models for recent mortality in palliative care, the best area under the curve (AUC) for predicting mortality is 0.68. Using the Modified Medical Research Council dyspnea score and oxygen saturation to define the combined dyspnea and oxygenation (DO) index, we find that the AUC of the DO index is 0.84 for predicting mortality in the validated cohort. Predictions of 1-year mortality based on the current palliative care guideline for COPD patients are poor. The DO index exhibits better predictive ability than other models in the study. npj Primary Care Respiratory Medicine (2022) 32:2 ; https://doi.org/10.1038/s41533-021-00263-7 INTRODUCTION developed to predict long-term survival. They all lacked accuracy 20–23 when applied to short-term events of <12 months . Marin The prevalence of and mortality associated with chronic et al. validated a number of existing prognostic indices in a large obstructive pulmonary disease (COPD) have been increasing individual pooled data set (n = 3633) from multiple cohort studies annually . However, current treatments have been disappointing with different stages of COPD. These prognostic indices included in terms of controlling airflow obstructions and reducing 2–4 the original BODE, the modified BODE (replacing the 6-min walk mortality . Although palliative care is shown to be effective in distance (6MWD) with peak oxygen uptake V’ as % predicted), O2 patients with COPD, these patients have fewer opportunities to 5,6 the BODEx (replacing the 6MWD with exacerbations), the eBODE receive palliative care than patients with cancer . Jabbarian et al. (BODE plus exacerbations), the SAFE (SGRQ score, air-flow found that the failure to implement advance care planning (ACP) limitation, and exercise tolerance), the ADO, and the DOSE. All- in chronic diseases is mainly due to the complexity and cause mortality prediction at 12 months was assessed for these unpredictability of the disease , and the uncertainty of disease indices, where the indices determined to be optimal for prediction 8–12 trajectory is even greater in COPD than in cancer . In addition, was the ADO (C statistic = 0.70). Boeck et al. developed the B-AE- COPD patients typically want to know more about their prognosis D indices (BMI, acute exacerbations, dyspnea) for 2-year mortality 13,14 in the early stages . Therefore, enormous effort has been made in the PROMISE study, and external validation of the B-AE-D was to find indicators to predict a poor prognosis accurately. performed in COCOMICS and the COMIC study for 1-year all-cause Researchers have found many indicators related to various mortality (C statistic = 0.68 and 0.74, respectively). Therefore, none adverse outcomes for COPD, including patient age, body mass of these indices had the strong predictive ability for 1-year index (BMI), dyspnea, smoking status, exercise capacity, acute mortality. In addition, none of these models were developed with 15–17 exacerbation, symptoms, and biological indicators . Unfortu- the specific aim of predicting all-cause mortality in stable COPD nately, as was the case with the first proposed indicator, FEV , patients within 12 months. there was no optimal way to predict mortality based on the 25 To the best of our knowledge, Bloom et al. was the only 17,18 indicator . After the multisystem involvement characteristic of research group to develop indicators (the BARC index) for COPD became known, the focus was moved to composite predicting 1-year mortality with the aim of palliative care in 15,19 indicators to achieve better predictive outcomes . The earliest advanced COPD (C statistic = 0.78 and 0.70 for the development developed and most widely investigated multicomponent indica- and validation cohorts, respectively). The variables in the BARC tors included the Body-Mass Index, Airflow Obstruction, Dyspnea, only required routinely collected non-specialist information, and Exercise Capacity (BODE) Index , which was also recom- which, therefore, helped identify patients seen in primary care mended for predicting outcomes by the Global Initiative for institutions, but a total of 18 variables were required. Because no Chronic Obstructive Lung Disease (GOLD) . Later, numerous existing indices had strong enough predictive ability for 1-year different indices were developed, including the Dyspnea and mortality in clinical practice, and very few indices were developed Airflow Obstruction (ADO) Index, the Dyspnea, Obstruction, with the specific aim of predicting 1-year mortality for palliative Smoking, Exacerbation (DOSE) Index, and various modifications care in stable COPD. In this study, we aimed to validate the of the BODE index . However, most of these indices were currently recommended prediction indices for palliative care, we Division of General Medicine, Department of Internal Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. 2 3 Department of Public Health, College of Medicine, National Cheng Kung University, Tainan, Taiwan. Division of Pulmonary Medicine, Department of Internal Medicine, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. Department of Internal Medicine and Oncology, National Cheng Kung University Hospital, College of Medicine, National Cheng Kung University, Tainan, Taiwan. These authors contributed equally: Tzuen-Ren Hsiue, Chiung-Zuei Chen. email: chen96@mail.ncku.edu.tw Published in partnership with Primary Care Respiratory Society UK 1234567890():,; S.-H. Tsai et al. also developed a new predictive index for 1-year mortality in Predictive variables for palliative care hospitalized ambulatory COPD patients. In the first part of the study, we evaluated the predictive ability of the currently recommended variables for estimating 1-year mortality in the palliative care guideline for COPD. We selected METHODS several variables for building the predictive model based on a 30–32 Study design review of the currently recommended prediction variables . We conducted this cohort study in the National Cheng Kung The variables included (1) mMRC score = 4, (2) frequent, severe AE (two or more AEs requiring hospitalization in the preceding year), University Hospital (NCKUH) from August 2006 to December 2015. (3) hypoxemia (SpO < 90% in ambient air), (4) BMI < 21, and (5) The patients included in the present study were part of another predicted FEV < 30%. We used several combined indices to test previous study . The patients were eligible for inclusion if they the accuracy of the prediction for 1-year mortality. The patients had received regular management for COPD at our hospital for >1 were subdivided into four groups: Group 1 was defined as patients year prior to their recruitment. All patients were diagnosed with with frequent, severe AE in combination with severe dyspnea COPD by pulmonologists according to the GOLD guidelines for (mMRC = 4). Group 2 was defined as patients with frequent, diagnostic criteria . The criteria were as follows: age >40 years, severe AE in combination with SpO < 90% in ambient air. Group 3 typical symptoms, such as cough, dyspnea, wheezing, or chest was defined as patients with frequent, severe AE combined with tightness in combination with evidence of chronic airflow predicted FEV < 30%. Group 4 was defined as patients with obstruction, as defined by a postbronchodilator ratio of forced frequent, severe AE in combination with BMI < 21. expiratory volume in 1 s (FEV ) to a forced vital capacity (FVC) of <70%. Pulmonary function tests were performed following the Modeling the predictive scores standard protocols of the American Thoracic Society . All patients Because of the generally unsatisfactory predictive power found in were enrolled under clinically stable conditions. We excluded previous studies and with validating our results, we wanted to patients who were unwilling to participate and those who had derive a new predictive model for 1-year mortality from the advanced lung cancer and pulmonary fibrosis because of patient variables, including age, sex, BMI, disease severity, such as anticipated death in the near future. Patients with missing data mMRC dyspnea score, FEV , SpO , and comorbidities. The 1 2 and those lost to follow-up in the first year were also excluded variables were evaluated using multivariate Cox regression models from the analysis. In total, 752 patients with COPD were analyzed with a forward entering approach and a 5% significance level for (Supplementary Fig. 1). The Institutional Review Board of NCKUH the selection criteria. Significant regression coefficients were approved this study before commencement (IRB number: B-ER- converted to exponential expressions for the weighting of the 105-386 and B-ER-98-289). Written informed consent was variables used for the predictive indices. obtained for all participants while enrollment. Validation of the predicting index Prognostic variables and outcome To validate the predictive performance of our model, we selected A total of 752 consecutive COPD patients were recruited. All a second cohort. All patients in the development group were patients were monitored through December 2016 or until death. recruited from pulmonary outpatient departments. Considering We acquired age, smoking history, BMI, the severity of dyspnea that if the validation group and the developmental group assessed by grade on the modified Medical Research Council exhibited high homogeneity, it was expected that the proposed model would obtain very similar results for the two groups of (mMRC) dyspnea scale , the degree of comorbidity as evaluated patients. Patients in the validation group were recruited by using the Charlson index , oxygen saturation levels as detected screening individuals who had been diagnosed with COPD, not by pulse oximetry in room air (SpO ), and status of long-term only in the pulmonary outpatient department but also in the home oxygen usage from every patient at the time of inclusion as Center for Hospice Palliative Shared Care at NCKUH from July 2012 determined by research assistants in the study. Comorbidity was to August 2019. All patients were aged ≥40 years; COPD was evaluated using the Charlson index and included congestive heart defined according to the GOLD diagnostic guidelines and criteria failure, coronary artery disease, systemic hypertension, peptic as the developmental group; patients with advanced lung cancer ulcer, and diabetes mellitus as identified from the patient files and or pulmonary fibrosis were excluded. The date of recruitment of detailed interviews. A severe acute exacerbation of COPD was some patients from the Center for Hospice Palliative Shared Care defined as an acute event characterized by a worsening of the overlapped with the time periods during which the development patient’s respiratory symptoms that were beyond day-to-day group was recruited. These patients were not excluded from this variations that also required hospitalization. The number of severe study since the source of patients was different from that for the exacerbations in the preceding year was recorded by research development group (Center for Hospice Palliative Shared Care assistants according to the patient’s chart as the primary means of versus the pulmonary outpatient department). All patients had data collection; self-reported data was used to supplement complete follow-up for 1 year or until death. this data. All-cause mortality was defined as the endpoint of the study. Statistical analysis The survival status of all patients was evaluated using a Continuous variables are presented as the median and inter- prospective observation, as reported in a previous study . All quartile range because the number of deaths was not large and patients were contacted during regular clinic visits or by therefore may not follow a normal distribution. Therefore, telephone interviews (if they missed an appointment). Most comparisons between survivors and nonsurvivors were performed patients who died during the study period had been regularly using Mann–Whitney U-test. Comparisons between categorical followed and had visited the hospital for treatment before their variables were performed using chi-square tests or Fisher’s death. Their dates of death were recorded and verified using exact tests. hospital records. Research assistants obtained the date of death of Kaplan–Meier survival curves and log-rank tests were used for patients who died outside the hospital by telephone contact with comparing different predictive variables. The ability to predict partners or family members. Survival status was also verified mortality within 1 year was analyzed using logistic regression through linkage with the Taiwan National Mortality Registry. models and the receiver operating characteristic (ROC) curve to npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK 1234567890():,; S.-H. Tsai et al. calculate the area under the curve (AUC). Data processing and severe AE and predicted FEV < 30% was 0.60, and the AUC for analyses were performed using the SPSS for Windows version patients with frequent, severe AE and BMI < 21 was 0.68. The ROC 17.0 statistical software (IBM, Armonk, NY, USA). for the different composite indices did not differ significantly between the four groups of combinations (Fig. 1). Kaplan–Meier survival analysis for all-cause mortality showed Reporting summary the 1-year survival rates for groups 1–4 were 62, 75, 88, and 78%, Further information on research design is available in the Nature respectively (Fig. 2). All indices, including those for groups 1–4, Research Reporting Summary linked to this article. showed high specificity but unsatisfactory sensitivity. The composite indices for group 1 with frequent, severe AE and RESULTS severe dyspnea had a better positive predictive value than the Participants other groups (Table 2). We enrolled 752 COPD patients from August 2006 through December 2015. The mean age of the patients was 70.6 years, and Development and validation of a prediction index most of them were men (92.7%). Twenty-eight percent of the In the univariate analysis, there were significant differences patients had severe to very severe airflow limitations (FEV < 50% between survivors and nonsurvivors in age, FEV , SpO , mMRC, of predicted); 25.2% of the patients had mMRC scores from 3 to 4, 1 2 BMI, anemia, and concurrent malignancies. Using multivariate and 50.7% had at least one severe AE in the preceding year during enrollment. At the end of the follow-up period in December 2016, regression, only mMRC and SpO were independent risk factors 378 patients had died (50.3%), and 60 patients (8%) had died for predicting 1-year mortality. within 1 year after the start of follow-up. We refined the dyspnea and oxygenation (DO) index by The baseline characteristics of survivors and nonsurvivors are weighting dyspnea and SpO based on the results of the shown in Table 1. Compared with survivors, nonsurvivors were multivariate regression model. We used the integers closest to older, had a worse pulmonary function, lower BMI, lower oxygen the hazard ratio for scoring predictive variables (Table 3). saturation, worse symptoms of dyspnea (higher mMRC score), and Compared with groups 1–4, the DO index had better discrimina- more AEs in the previous year. tion for mortality (AUC = 0.73; Fig. 1). We also performed a sensitivity analysis for patients with severe or very severe Predictive ability of the currently recommended models obstruction (FEV < 50%), where the DO index performed better The AUC values for predicting 1-year mortality in patients with (AUC = 0.81). The survival rates for the different DO scores are severe dyspnea (mMRC = 4) and patients having frequent, severe AE were 0.62 and 0.60, respectively. Combining predictor variables for patients with severe dyspnea and frequent severe AE was better than using only one variable (AUC = 0.68). The AUC for predicting 1-year mortality for patients with frequent, severe AE and SpO < 90% was 0.66. The AUC for patients with frequent, Table 1. Demographic and patient characteristics of survivors and nonsurvivors. Characteristic Survivors Nonsurvivors p Value (n = 692) (n = 60) Age, median (IQR) 71.2 (64.6, 78.7) 78.4 (72.5, 81.6) <0.01 Male n (%) 640 (92.5) 57 (95.0) 0.61 Fig. 1 ROC curve for severe dyspnea, severe acute exacerbation, Current smoker, 189 (27.3) 13 (21.7) 0.30 and different combinations of predictors and DO index for 1-year n (%) mortality in COPD patients. The AUC values for patients with severe Smoking quantity 45 (23, 70) 50 (20, 62) 0.92 dyspnea (mMRC = 4), frequent severe AE, groups 1–4, and DO index (pack-years) were 0.62, 0.60, 0.68, 0.66, 0.60, 0.68, and 0.73, respectively. FEV % 64 (48, 82) 50 (34, 64) <0.01 BMI 23.3 (20.5, 25.8) 20.5 (17.0, 24.5) <0.01 SpO % 97.0 (95.0, 98.0) 95.5 (92.0, 97.0) <0.01 CI score 2.0 (1.0, 3.0) 3.0 (1.0, 5.0) <0.01 Severe AE ≥ 2, n (%) 111 (16.0) 22 (36.7) <0.01 6MWT (meter) 344.0 (248.0, 400.0) 278.0 (206.0, 313.0) 0.11 SGRQ score 33.22 (18.2, 51.0) 59.13 (46.4, 65.3) <0.01 mMRC= 4 26 (3.7) 17 (28.3) <0.01 LTOT, n (%) 71 (10.3) 15 (25.0) <0.01 Discrete data are presented as number (percentage), and continuous variables are presented as median (IRQ). FEV forced expiratory volume in 1 s, BMI body mass index, SpO oxygen 1 2 saturation (%) detected with pulse oximeter when breathing in room air, CI Charlson index, severe AE ≥ 2 history more than one acute exacerbation that Fig. 2 Kaplan–Meier survival curves for 1-year mortality accord- required hospitalization in the preceding year, 6MWT 6 min walking test, ing to different recommended prediction indices for palliative SGRQ St. George’s Respiratory Questionnaire, mMRC modified Medical care. The 1-year survival rates were 62, 75, 88, and 78% for groups Research Council Dyspnea Scale, LTOT long-term oxygen therapy. 1–4, respectively. Published in partnership with Primary Care Respiratory Society UK npj Primary Care Respiratory Medicine (2022) 2 S.-H. Tsai et al. Table 2. Predictive accuracy of different recommended palliative care indices for 1-year mortality. Prognostic index Sensitivity Specificity PPV NPV Accuracy AUC mMRC= 4 28.3% 96.2% 39.5% 93.9% 90.8% 0.623 Severe AE ≥ 2 36.7% 84.0% 16.5% 93.9% 80.2% 0.603 Group 1 13.3% 98.1% 38.1% 92.9% 91.4% 0.684 Group 2 3.3% 99.1% 25.0% 92.2% 91.5% 0.657 Group 3 3.3% 98.0% 12.5% 92.1% 90.4% 0.634 Group 4 18.3% 94.4% 22.0% 93.0% 88.9% 0.679 mMRC modified Medical Research Council Dyspnea Scale in stable condition, Severe AE ≥ 2 more than one acute exacerbation that required hospitalization in the preceding year, Group 1 mMRC = 4 + severe AE ≥ 2, Group 2 severe AE ≥ 2 + SpO < 90%, Group 3 severe AE ≥ 2 + FEV < 30%, Group 4 severe AE ≥ 2 + BMI 2 1 < 21, PPV positive predictive value, NPV negative predictive value, AUC area under the curve, SpO oxygen saturation (%) detected with a pulse oximeter when breathing room air, FEV forced expiratory volume in 1 s, BMI body mass index. Table 3. Weighting of variables in DO index. Table 4. Survival analysis of 1-year mortality for different DO index scores of patients with severe and very severe COPD (n = 180). Variable β Adjusted HR Score Score Survived (n) Died (n) Survival Statistics p Value SpO (%) 2 rate (%) (chi-square) 95–100 0 1 1 DO 58.61 <0.001 90–94 0.55 1.7 2 2 56 2 96.55 85–89 1.05 2.9 3 3 58 2 96.67 <85 2.00 7.3 7 4 30 3 90.91 mMRC score 5 2 0 100.00 0–20 1 1 8 1 0 100.00 3 0.64 1.9 2 90 1 0 4 2.21 9.1 9 10 10 2 83.3 Coding according to the regression coefficient for DO index construction. 11 4 4 50 DO dyspnea and oxygenation. 12 1 2 33.3 16 0 2 0 shown in Table 4. In the group with the most severe COPD Score Survived (n) Died (n) Survival Statistics p Value (DO score = 12–16), the survival rate was only 20%. rate (%) (chi-square) We enrolled a total of 342 patients for the validation group. The patients in the validation group were older (73.5 vs. 72.2%), had DO 52.98 <0.001 lower oxygenation (96 vs. 97%), more comorbidities, as evaluated 2–7 147 7 95.45 by the CI score (4.0 vs. 2.0%), more symptoms, as evaluated by the 8–11 14 7 66.7 percentage of patients with mMRC = 4 (14.3 vs. 6.3%), and higher 1-year mortality (14.3 vs. 8.0%), than the developmental group 12–16 1 4 20.00 (Supplementary Table 1). When applying our DO score for Wilcoxon test. predicting 1-year mortality, the AUC was 0.84. In the group with b Log-rank test. the most severe COPD (DO score = 12–16), the positive and negative predictive values were 87% and 89%, respectively, for 1-year mortality. guideline-recommended prediction tool. There was no process for sample size calculation in the study. Therefore, we calculated the statistical power backward using our sample size. According to DISCUSSION 23–25 the predictive ability of existing indices in a previous review , This study showed that a combination of mMRC and frequent, we considered an AUC of 0.65 as the median discrimination power severe AE as a predictor of 1-year mortality demonstrated similar for previous predictors. Using our population of a total of 342 poor discrimination power as other combinations of factors. These patients in the validation group for a two-sided z-test at a factors included predictors with desaturation, or a poor grade of significance level of 0.05, we achieved a 99% power to detect the lung function, or low BMI combined with frequent, severe AE. The AUC between the median discrimination power of previous AUC values of these combinations ranged from 0.60 to 0.68. In predictors and the DO index in this study (PASS Power Analysis addition, all combined groups exhibited lower sensitivity but and Sample Size Software, NCSS, LLC., Kaysville, Utah). However, higher specificity. Therefore, these indices did a good job of ruling there were only 18 deaths out of 180 patients when the DO index outpatients who would survive for >1 year but tended to miss was applied in severe COPD patients (see Table 4). The small patients who would die within 1 year. numbers in some lattices may have thus affected the accuracy of Using the mMRC dyspnea score and oxygen saturation (SpO ), we developed a better discriminating model for 1-year mortality, the estimate. For example, when using a DO score = 9 as a cutoff value, the 1-year survival rate for patients with DO scores ≥ 9 was the DO index. The AUC value of the DO index was 0.73 for the prediction of 1-year mortality, and the AUC was 0.84 in the estimated to be 58% (15/26). The accuracy of the predictive ability, validated cohort, which was superior to the current palliative sensitivity, specificity, positive predictive value, and negative npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK S.-H. Tsai et al. predictive value was 87.8% (95% CI (confidence interval), and spirometry results, and the deaths were verified by linking 82.1–92.2%), 61.1% (95% CI, 35.8–82.7), 90.7% (95% CI, with a database from the Taiwan National Mortality Registry. In 85.2–94.7), 42.3% (95% CI, 28.6–57.4), and 95.5% (95% CI, addition, we performed a sensitivity analysis by excluding a small 92.2–97.4), respectively. The low precision for sensitivity and the amount of missing mMRC values and by restricting the subjects to positive predictive value as indicated by wider confidence patients with severe airflow limitations. Both results showed intervals was attributed to the smaller sample size in this test. similar findings, which indicated the consistency of the discrimi- In general, the discriminative power of the model for the native validity. Finally, we validated the DO index and showed its development group was better than that for the validation group. good predictive ability for 1-year mortality in the second cohort of Patients in the validation group were older, had lower oxygena- 342 patients. tion, more comorbidities, more symptoms, and higher mortality There are limitations to our study. First, we did not have a large than the development group in this study. The heterogeneity patient population and were only limited to one medical center. between our developmental and validation groups was one of the The second limitation of the present study was that few women reasons explaining why the predictive ability was better in the (7.3%) were enrolled. Recently, a large, real-world, cohort study validation group than in the development group. To evaluate revealed gender differences among COPD patients, where COPD whether the predictive ability of the DO index becomes stronger was more frequent among women (53.8%), and the overall over time, we analyzed the predictive ability for 3- and 5-year mortality rate was higher in men as compared with women (45 vs. mortality and found that the AUC values were 0.66 and 0.67, 38%). However, no differences in mortality due to COPD related to respectively. These results implied that the DO index is not gender were found . The majority of participants with smoking- suitable for predicting 3- and 5-year mortality. related COPD in Taiwan are male ; the smoking prevalence for The findings of poor discrimination for the current models in women is <5% in Taiwan. In contrast to industrialized countries in this study were consistent with a previous systemic review the West, COPD morbidity remains male predominant in most 33 37–39 conducted by Almagro et al. . They used indicators already Asian countries . However, the result of this study cannot be developed in previous articles to validate the performance in their directly generalizable to other countries due to this limitation. cohort. Composite indices were better than a single parameter, The third limitation was that socioeconomic status (SES) was not and the best AUC was 0.68 for the CODEX index (comorbidity, included in our predictive model. SES disadvantages appear to obstruction, dyspnea, and previous exacerbation). The author have a significant impact on COPD mortality and morbidity, where concluded that no single index is good enough to guide the individuals with the lowest SES consistently had been shown to initiation of palliative care. Thus, the clinician should not make this have significantly higher mortality than those with the highest 40,41 decision based solely on a predictive tool. However, the use of the SES . We thus suggest including an SES measurement in the proposed DO index improved the predictive power (AUC = 0.73) predictive model in further studies. for 1-year mortality, and the AUC was 0.84 in the validated cohort. In conclusion, this study demonstrated that the predictive The BARC index for prognostic factors, including BMI and blood values for 1-year mortality were poor, based on the current results (B), age (A), respiratory variables (airflow obstruction, recommendations for palliative care among COPD patients, exacerbations, smoking) (R), and comorbidities (C) was conducted including four different composite indices. The newly developed based on medical databases and had a satisfactory AUC for 1-year DO index proposed in this work exhibited better predictive ability mortality (AUC = 0.79). However, it included 18 variables in the than other alternatives. We suggest that COPD patients with DO model, such as age, BMI, FEV , severe exacerbations, smoking index scores ≥12, for example, patients with mMRC = 4 and SpO 1 2 status, multiple comorbidities, hemoglobin, platelets, and others < 90%, are good candidates to receive palliative care. for the evaluation . In contrast, the DO index proposed in this study is simple to use because only two clinical parameters, DATA AVAILABILITY dyspnea score and oxygenation, detected with a pulse oximeter, Full data sets are not available publicly currently for protecting patient privacy. But are needed. Another composite index, the ProPal-COPD tool, had the data can be requested reasonably to the corresponding author. a good predictive ability for 1-year mortality with an AUC of 0.82. This model relied on the following seven predictors: (1) a surprise question, (2) MRC dyspnea, (3) the Clinical COPD Questionnaire Received: 6 November 2020; Accepted: 25 November 2021; (CCQ), (4) FEV % of predicted value, (5) BMI, (6) previous hospitalizations for AECOPD, and (7) specific comorbidities. However, some variables, such as the surprise question and CCQ, were not always routinely captured. REFERENCES The DO index developed in this study was composed of the 1. Rabe, K. F. et al. Global strategy for the diagnosis, management, and prevention mMRC score and oxygen saturation. In a recent systematic review of chronic obstructive pulmonary disease: GOLD executive summary. Am. J. of predictive indicators in COPD, 24 models used composite Respir. Crit. Care Med. 176, 532–555 (2007). indicators . The dyspnea score (mMRC) is one of the ten most 2. Tashkin, D. P. et al. A 4-year trial of tiotropium in chronic obstructive pulmonary used parameters. Nishimura et al. also demonstrated that the disease. N. Engl. J. Med. 359, 1543–1554 (2008). severity of dyspnea was a more favorable predictor of death than 3. Yang, I. A., Fong, K. M., Sim, E. H., Black, P. N. & Lasserson, T. J. Inhaled corti- FEV . This result was consistent with the use of mMRC in this costeroids for stable chronic obstructive pulmonary disease. Cochrane Database article. Our model also included another indicator, SpO , which Syst. Rev. CD002991 (2007). was not commonly used in previous studies. Instead, some studies 4. Calverley, P. M. et al. Salmeterol and fluticasone propionate and survival in used arterial oxygen partial pressure (PaO ) as a predictor. PaO is chronic obstructive pulmonary disease. N. Engl. J. Med. 356, 775–789 (2007). 2 2 5. Au, D. H., Udris, E. M., Fihn, S. D., McDonell, M. B. & Curtis, J. R. Differences in most often used during hospitalization, where PaO can fluctuate health care utilization at the end of life among patients with chronic obstructive due to many factors, such as oxygen use, pneumonia, or pulmonary disease and patients with lung cancer. Arch. Intern. Med. 166, 326–331 cardiovascular instability. Additionally, drawing arterial blood (2006). may also lead to some local complications. In our study, we used 6. Jabbarian, L. J. et al. Advance care planning for patients with chronic respiratory SpO in stable patients breathing ambient air to measure constant diseases: a systematic review of preferences and practices. Thorax 73, 222–230 oxygenation status. Lower levels of invasiveness are also preferred (2018). in outpatient settings. 7. Higginson, I. J. et al. An integrated palliative and respiratory care service for The strength of this study is its diagnostic and measurement patients with advanced disease and refractory breathlessness: a randomized accuracy. COPD was diagnosed according to standard evaluations controlled trial. Lancet Respir. Med. 2, 979–987 (2014). Published in partnership with Primary Care Respiratory Society UK npj Primary Care Respiratory Medicine (2022) 2 S.-H. Tsai et al. 8. Crawford, A. Respiratory practitioners’ experience of end-of-life discussions in 35. Bellou, V., Belbasis, L., Konstantinidis, A. K., Tzoulaki, I. & Evangelou, E. Prognostic COPD. Br. J. Nurs. 19, 1164–1169 (2010). models for outcome prediction in patients with chronic obstructive pulmonary 9. Gott, M. et al. Barriers to advance care planning in chronic obstructive pulmonary disease: systematic review and critical appraisal. BMJ https://doi.org/10.1136/bmj. disease. Palliat. Med. 23, 642–648 (2009). l5358 (2019). 10. Scheerens, C. et al. “A palliative end-stage COPD patient does not exist”: a qua- 36. Lisspers, K. et al. Gender differences among Swedish COPD patients: results from litative study of barriers to and facilitators for early integration of palliative home the ARCTIC, a real-world retrospective cohort study. NPJ Prim. Care Respir. Med. care for end-stage COPD. NPJ Prim. Care Respir. Med. https://doi.org/10.1038/ https://doi.org/10.1038/s41533-019-0157-3 (2019) s41533-018-0091-9 (2018). 37. Jamrozik, E. & Musk, A. W. Respiratory health issues in the Asia-Pacific region: an 11. Gaspar, C., Alfarroba, S., Telo, L., Gomes, C. & Barbara, C. End-of-life care in COPD: overview. Respirology 16,3–12 (2011). a survey carried out with Portuguese pulmonologists. Rev. Port. Pneumol. 20, 38. Tan, W. C. & Ng, T. P. COPD in Asia: where East meets West. Chest 133, 517–527 123–130 (2014). (2008). 12. Smith, T. A. et al. Specialist respiratory physicians’ attitudes to and practice of 39. Tan, W. C. Trends in chronic obstructive pulmonary disease in the Asia-Pacific advance care planning in COPD. A pilot study. Respir. Med. 108, 935–939 regions. Curr. Opin. Pulm. Med. 17,56–61 (2011). (2014). 40. Gershon, A. S., Dolmage, T. E., Stephenson, A. & Jackson, B. Chronic obstructive 13. MacPherson, A., Walshe, C., O’Donnell, V. & Vyas, A. The views of patients with pulmonary disease and socioeconomic status: a systematic review. COPD 9, severe chronic obstructive pulmonary disease on advance care planning: a 216–226 (2012). qualitative study. Palliat. Med. 27, 265–272 (2013). 41. Sahni, S., Talwar, A., Khanijo, S. & Talwar, A. Socioeconomic status and its rela- 14. Janssen, D. J. A., Spruit, M. A., Schols, J. M. G. A. & Wouters, E. F. M. A call for high- tionship to chronic respiratory disease. Adv. Respir. Med. 85,97–108 (2017). quality advance care planning in outpatients with severe COPD or chronic heart failure. Chest 139, 1081–1088 (2011). 15. Dijk, W. D. et al. Multidimensional prognostic indices for use in COPD patient ACKNOWLEDGEMENTS care. A systematic review. Respir. Res. https://doi.org/10.1186/1465-9921-12-151 The study was funded by grants from the Ministry of Science and Technology (2011). (MOST107-2627-M-006-007, MOST 109-2314-B-006-091, MOST 110-2314-B-006-099), 16. Man, S. F. et al. C‐reactive protein and mortality in mild to moderate chronic and it was supported in part by the Higher Education Sprout Project, Ministry of obstructive pulmonary disease. Thorax 61, 849–853 (2006). Education to the Headquarters of University Advancement at National Cheng Kung 17. Nishimura, K., Izumi, T., Tsukino, M. & Oga, T. Dyspnea is a better predictor of University (NCKU). We are grateful to the Health and Welfare Data Science Center of 5-year survival than airway obstruction in patients with COPD. Chest 121, the Ministry of Health and Welfare and the Center of hospice palliative shared care at 1434–1440 (2002). National Cheng Kung University Hospital (NCKUH) for providing all the data sets, 18. Casanova, C. et al. Differential effect of modified medical research council dys- facilities, and linkage services that were required for this study. Parts of our results pnea, COPD assessment test, and clinical COPD questionnaire for symptoms were submitted as an e-poster in the ATS 2020 Virtual content. evaluation within the new GOLD staging and mortality in COPD. Chest 148, 159–168 (2015). 19. Celli, B. R. Predictors of mortality in COPD. Respir. Med. 104, 773–779 (2010). AUTHOR CONTRIBUTIONS 20. Celli, B. R. et al. The body-mass index, airflow obstruction, dyspnea, and exercise T.-R.H. and C.-Z.C. were involved in the concept development and study design. All capacity index in chronic obstructive pulmonary disease. N. Engl. J. Med. 350, authors contributed to the data retrieval and patient recruitment. S.-H.T. and C.-Y.S. 1005–1012 (2004). conducted the analysis. Paper drafting was mainly accomplished by S.-H.T. and C.-Z.C. 21. Vestbo, J. et al. Global strategy for the diagnosis, management, and prevention of and all authors contributed to the revision and approval of the final manuscript. chronic obstructive pulmonary disease: GOLD executive summary. Am. J. Respir. Crit. Care Med. 187, 347–365 (2013). 22. Smith, L. E. et al. Prognostic variables and scores identifying the end of life in COMPETING INTERESTS COPD: a systematic review. Int. J. Chron. Obstruct. Pulmon. Dis. 12, 2239–2256 (2013). The authors declare no competing interests. 23. Marin, J. M. et al. Multicomponent indices to predict survival in COPD: the COCOMICS study. Eur. Respir. J. 42, 323–332 (2013). 24. Boeck, L. et al. Prognostic assessment in COPD without lung function: the B-AE-D ADDITIONAL INFORMATION indices. Eur. Respir. J. 47, 1635–1644 (2016). Supplementary information The online version contains supplementary material 25. Bloom, C. I., Ricciardi, F., Smeeth, L., Stone, P. & Quint, J. K. Predicting COPD 1-year available at https://doi.org/10.1038/s41533-021-00263-7. mortality using prognostic predictors routinely measured in primary care. BMC Med. https://doi.org/10.1186/s12916-019-1310-0 (2019). Correspondence and requests for materials should be addressed to Chiung-Zuei 26. Chen, C. Z. et al. Using post-bronchodilator FEV1 is better than pre- Chen. bronchodilator FEV1 in evaluation of COPD severity. COPD 9, 276–280 (2012). 27. Blonshine, S., Mottram, C. D. & Wanger, J. Pulmonary Function Laboratory Man- Reprints and permission information is available at http://www.nature.com/ agement and Procedure Manual 2nd edn (American Thoracic Society, 2005). reprints 28. Bestall, J. C. et al. Usefulness of the medical research council (MRC) dyspnea scale as a measure of disability in patients with chronic obstructive pulmonary disease. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims Thorax 54, 581–586 (1999). in published maps and institutional affiliations. 29. Charlson, M. E., Pompei, P., Ales, K. L. & MacKenzie, C. R. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J. Chronic Dis. 40, 373–383 (1987). 30. Lanken, P. N. et al. ATS End-of-Life Care Task Force. An official American Thoracic Open Access This article is licensed under a Creative Commons Society clinical policy statement: palliative care for patients with respiratory Attribution 4.0 International License, which permits use, sharing, diseases and critical illnesses. Am. J. Respir. Crit. Care Med. 177, 912–927 (2008). adaptation, distribution and reproduction in any medium or format, as long as you give 31. O'Donnell, D. E. et al. Canadian Thoracic Society recommendations for man- appropriate credit to the original author(s) and the source, provide a link to the Creative agement of chronic obstructive pulmonary disease - 2008 update. Can. Respir. J. Commons license, and indicate if changes were made. The images or other third party 15,1A–8A (2008). material in this article are included in the article’s Creative Commons license, unless 32. Thomas, K., Wilson, J. A. & GSF Team. GSF PIG 6th Edition. National Gold Stan- indicated otherwise in a credit line to the material. If material is not included in the dards Framework Centre in end of life care. http://www.goldstandardsframework. article’s Creative Commons license and your intended use is not permitted by statutory org.uk (2016) regulation or exceeds the permitted use, you will need to obtain permission directly 33. Almagro, P. et al. Palliative care and prognosis in COPD: a systematic review with from the copyright holder. To view a copy of this license, visit http://creativecommons. a validation cohort. Int. J. Chron. Obstruct. Pulmon. Dis. 12, 1721–1729 (2017). org/licenses/by/4.0/. 34. Yohannes, A. M., Baldwin, R. C. & Connolly, M. J. Predictors of 1-year mortality in patients discharged from hospital following acute exacerbation of chronic obstructive pulmonary disease. Age Ageing 34, 491–496 (2005). © The Author(s) 2022 npj Primary Care Respiratory Medicine (2022) 2 Published in partnership with Primary Care Respiratory Society UK

Journal

npj Primary Care Respiratory MedicineSpringer Journals

Published: Jan 13, 2022

References