Chronic obstructive pulmonary disease (COPD) is the seventh leading cause of poor health and the third leading mortality disease worldwide.1 Despite increasing awareness of COPD burden, COPD remains highly underdiagnosed worldwide, ranging from 65% to 80% in the community,2 and a substantially high proportion of underdiagnosis occurs in primary care (PC).3–5 Underuse of spirometry is the strongest predictor for underdiagnosis.6 Low symptom perception and insufficient awareness of risk factors by the patients and their clinicians are also important contributing factors.2,7 Additionally, a large-scale study (n=30,874) by Lamprecht et al. discovered that COPD underdiagnosis was also associated with male gender, younger age, never and current smoking, lower education, no previous spirometry, and less severe airflow limitation.6 Hence, an effective COPD case-finding strategy to identify at-risk persons for confirmatory spirometry is urgently needed.

Currently, screening for COPD in asymptomatic adults is not recommended by the US Preventive Services Task Force.8,9 However, case-finding for early COPD identification in those with respiratory symptoms or exposure risks is advocated.10 The latter often present to healthcare facilities, particularly in PC, where are the optimal places to catch COPD early. Case-finding tools in PC include questionnaires, handheld devices, or a combination of both.11,12 Symptom-based questionnaires, such as the COPD diagnostic questionnaire,13,14 COPD Population Screener questionnaire (COPD-PS),15,16 and lung function questionnaire,17 are commonly used tools. These questionnaires exerted a poor to fair diagnostic accuracy (DA, indicated by the area under the receiver operating characteristic curve [AUROC]=0.64–0.71), a wide range of sensitivity (63–91%), and specificity (45–70%). Compared with questionnaires, case-finding tools using handheld devices, such as simplified spirometers,14,16 or peak expiratory flow meters18,19 exhibited better DA (AUROC=0.80–0.88), similar sensitivity (76–88%) and higher specificity (72–95%). Additionally, combined questionnaires and handheld devices improved DA (AUROC=0.87–0.91).14,18,19 Taken together, questionnaires are simple and sensitive but less specific case-finding tools. A questionnaire needs linguistic validation before being widely used in people who speak different languages. Applying handheld devices can elevate specificity and reduce the numbers needed to screen (NNS). Different countries should establish their own case-finding strategies.

Recently, the English PUMA questionnaire (7 items, score 0–9, higher scores indicating a higher diagnostic rate) has emerged as a new effective COPD case-finding tool with acceptable DA and predictive performance (AUROC=0.76, sensitivity/specificity/NNS=74%/65%/4 at the cutoff score ≥5) in Latin American countries.20 Additionally, Au-Doung et al. reported that a translated Chinese PUMA version (for Cantonese) exhibited similar performance (AUROC=0.753; sensitivity/specificity=77%/63% at the cutoff score ≥6) in Hong Kong.21 In Taiwan, a recent nationwide telephone interview survey of the general population revealed that COPD was largely underdiagnosed.22 Taiwan lacks an effective, validated Chinese questionnaire for early identification of COPD so far. Thus, Chinese PUMA (for Taiwanese) can be a feasible solution. Additionally, monitoring peak expiratory flow rate (PEFR) is widely used in PC for asthma control in Taiwan. The present study's primary objective was to form a linguistically validated Chinese PUMA (C-PUMA, Mandarin version) and examine its DA with or without the combination of PEFR in Taiwan. The other objectives included the reliability and validity of C-PUMA and the predictive performance of different diagnostic modalities.

The present study applied the standard forward–backward translation processes to form the C-PUMA, followed by confirmation of reliability, validity, and clinical performance. C-PUMA alone showed fair DA, which was enhanced by the PUMA-PEFR prediction model. These tools might act as effective COPD case-finding tools in PC.

Linguistic validation and psychometric evaluation are necessary before a translated questionnaire is widely used in the targeted population. Although another Chinese PUMA questionnaire (HK-PUMA, for Cantonese) was reported by Au-Doung PLW et al. in Hong Kong,21 the details of the validation process were not reported. In the present study, excellent test–retest reproducibility (Table 1) indicates that C-PUMA measures a non-biased concept over time, and moderate-to-high internal consistency reflects that each item of C-PUMA constantly links to COPD diagnosis. This process ensured that C-PUMA was qualified before use in clinical validation. Additionally, C-PUMA exhibited weak but significant association with CAT (concurrent validity) and SF-12V2-PCS (predictive validity), indicating C-PUMA potentially links to COPD-specific symptom burdens and generic physical health status, respectively. Therefore, the translated C-PUMA is well-linguistically validated and suitable for use in the Chinese population.

Currently, PUMA has been reported in Latin American countries (original and externally validated English version),20,33 and in Hong Kong (Chinese validation for Cantonese).21 Compared with these PUMA studies (see the details in Table A3, Appendix A), our study had a similar study design except for the two major differences – respiratory symptoms and study settings. It might be concerned that in addition to age and smoke (tobacco or biomass) exposure, this study added respiratory symptoms to the inclusion criteria. The previous PUMA studies usually recruited subjects during routine spontaneous or scheduled appointments unrelated to the study in PC, irrespective of respiratory symptoms. However, the US Preventive Services Task Force recommended against screening for COPD in asymptomatic adults in 2016,8 which was reaffirmed in 2022.9 In contrast, the GOLD guideline recommended case-finding in symptomatic patients.10 The present study was conducted in a medical center, and patients seldom presented to our pulmonologist clinics without respiratory symptoms. Moreover, these symptomatic patients might benefit from timely diagnosis and treatment. This concept was proved in a very recent multicenter, randomized, controlled trial reported by Aaron et al.34 They used a case-finding method to identify adults in the community with respiratory symptoms without a previous diagnosis of COPD or asthma. Patients with confirmed diagnosis was assigned to receive either pulmonologist-directed treatment or usual care for 1 year. They found that the interventional group had less healthcare utilization for respiratory illness in comparison with the usual-care group. Therefore, respiratory symptoms might play an important role while initiating a case-finding strategy for early COPD detection, and these case-finding tool-detected patients had beneficial health outcomes from pulmonologist care.

The other difference is clinical settings. The impact of clinical settings on the case-finding efficacy of any specific questionnaire has not been reported to date. Interestingly, the DA of our hospital-based PUMA study (AUROC 0.749) was similar to that in previously PC-based (AUROC 0.704–0.753) or population-based (AUROC 0.734) studies (Table A3). As to predictive performance, in terms of the best cutoff score/sensitivity/specificity, our data (≥6/77%/64%) are almost identical to that in the HK-PUMA cohort (≥6/76.5%/63.3%).21 However, there are variations in sensitivity (51.5–77%), specificity (62.1–81.6%), PPV (39.1–59.9%), and NPV (63.3–88%) across these PUMA studies (Table A3). Different clinical settings and COPD incidence (18.7–45.1%) are the possible contributing factors. Collectively, PUMA exhibited similar DA but variable performance, indicating PUMA can be reliably used in English and Chinese populations, but different countries/regions must establish their own cutoff value before being widely used.

In the present study, the best C-PUMA cutoff score is ≥6, which had a more cost-effective NNS of 3 (vs. ≥5, NNS=4), but at a cost of missing more COPD patients up to 11.5%. Recently, CH Lin et al. reviewed different case-finding approaches and found that COPD incidence in PC ranged from 10% to 26% in those aged ≥40 years with a smoking history.11 Considering the high incidence of COPD in at-risk patients (e.g., 32% in the present study and 48.8% in our previous report18) and large underdiagnosis in the general population in Taiwan,22 we suggest the C-PUMA cutoff score ≥5 is appropriate for generalized use in PC to reduce COPD underdiagnosis if applying C-PUMA alone. Similarly, the HK-PUMA study also proposed that a cutoff score ≥5, which elevated sensitivity (91.2%) and NPV (92.7%), was more suitable than the Youden index-determined cutoff score ≥6 for their early detection strategy.21 Interestingly, the best English PUMA cutoff score also varied between ≥5 and ≥6 (Table A3).20,33 This might be speculated based on our data that a substantial and equal portion of COPD and non-COPD cases had their scores overlapped between this range. That is, the cutoff score of 5 or 6 has limited discrimination between COPD and non-COPD cases. Therefore, a secondary tool, such as PEFR, can enhance case discrimination in a case-finding approach.

This study found that PEFR was more accurate than C-PUMA, which was consistent with the recent studies that reported PEFR was also better than different questionnaires.18,19,35 However, the cutoff values of PEFR might largely vary across different studies, such as %PEFR <79%18 or <80%,36 PEFR <350, or <250L/min (for males and females, respectively).19 By contrast, COPD-PS was found to be more accurate than PEFR (cutoff value ≤2.2l/sm2).37 These discrepancies might be ascribed to different peak flow meters, clinical settings (PC vs. hospital), and various COPD incidences (17.4–49%) (summarized in Table A4, Appendix A). Additionally, Perez-Padilla et al. reported that %PEFR ≥70% could effectively exclude those with severe airflow limitation (FEV1 <50% predicted).38 Jithoo et al. suggested using PEFR with appropriate cutoff values could effectively reduce confirmatory spirometry numbers.39 Take together, PEFR is a useful case-finding tool but needs validation accordingly.

Combined various questionnaires and PEFR has been reported to serve as effective case-finding approaches in different study designs. In Spain, Soriano et al. reported that a combined modality selecting those with COPD-PS score ≥4 and pre-bronchodilation PEFR ≤2.2l/sm2 exhibited higher DA (AUROC 0.761) than that with COPD-PS (AUROC 0.715) or PEFR (AUROC 0.664) alone. This combined modality can reduce 90% of spirometry tests.37 In the US, Martinez et al. developed the novel CAPTURE questionnaire and classified the risk levels by scores (low: 0–1; middle: 2–4; high: 5–6) or by PEFR (high: males <350L/min, females <250L/min) for identification of clinically significant COPD (those with predicted FEV1 <60% and/or exacerbation risk).19 They reported that the combined modality resulted in the best DA (AUROC 0.9057), followed by PEFR (AUROC 0.8783) and CAPTURE (AUROC 0.7954). They concluded that those with middle scores should measure PEFR and those with high scores or middle scores with low PEFR should undergo spirometry. Identical case-finding protocol using the translated Spanish CAPTURE plus PEFR in those who spoke Spanish replicated identical results.40 Following these case-finding approaches, clinicians should independently evaluate two separate parameters (questionnaire scores and PEFR) to determine who should be subjected to confirmatory spirometry. This approach is likely to miss COPD patients who lack symptom perception (less symptom scores, such as case 2 in Table 4) or have preserved or mildly impaired lung function (less PEFR reduction, such as case 4 in Table 4). A prediction model possibly minimizes this risk. A prediction model can incorporate different factors into a single number, which number represents an integrated evaluation. Various COPD prediction models have been used to predict the risks of underdiagnosis41,42 and the probability of diagnosis.18 Recently, we incorporated age, CAT scores, smoking pack-years, and %PEFR to predict PCOPD.18 This integrated model exhibited the best DA (AUROC 0.866), compared to CAT (AUROC 0.666) or % PEFR (AUROC 0.832) alone. In this study, the integrated PUMA-PEFR model exhibited the best DA again. However, regarding combined questionnaires and PEFR, the evidence of whether a prediction model exerts better case-finding efficacy than evaluating two independent variables is lacking. Clinicians can develop their own combined modality based on available tools.

The strength of this study is that C-PUMA was well-translated and validated before use in clinical validation. There are some limitations. First, the data were driven in a medical center, not in PC. Actually, the healthcare system in Taiwan lacks mandatory referral regulations, and any outpatient has the freedom to visit any specialist in any hospital without a referral.43 Therefore, in the present study, 90.8% of participants came from the community, that is, our data will be very similar to those generated in PC. Nevertheless, validation in genuine PC is required to expand its generalizability. Second, a substantial portion of newly diagnosed COPD patients had mild disease (group A 77% or stage I 47%). Although new evidence has shown that case-finding tool-detected patients might benefit from specialist care,34 the optimal medication choice has not been well-established.

Our data are in line with the conclusions of a recent review by Aaron SD et al. that the use of combined tools has shown better diagnostic accuracy than either tool alone and that the combined use of a questionnaire and a handheld device is a more effective strategy for identifying individuals at increased risk for COPD, although questionnaires used alone are still valuable tools to predict COPD.12 In Taiwan, C-PUMA is well-linguistically and clinically validated. The best cutoff score is ≥6. However, the cutoff score ≥5 is considered appropriate to detect more undiagnosed COPD patients if applying C-PUMA alone. Furthermore, the PUMA-PEFR prediction model exhibits the best DA. Applying these new tools in PC can provide an opportunity to reduce COPD underdiagnosis.

This study was approved by the Institutional Review Board of Taipei Veterans General Hospital (approval number: 2019-05-009CC and 2019-08-007AC). All participants signed informed consent.

This work was supported by AstraZeneca Inc. The funder had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conception and design: D.W. Perng, K.C. Su, Y.H. Hsiao; grant application: D.W. Perng and K.C. Su; acquisition, analysis, or interpretation of data: K.C. Su, H.K. Ko, K.T. Chou, Y.H. Hsiao, and T.H. Jeng; drafting of the manuscript: K.C. Su, H.K. Ko, Chou, Y.H. Hsiao, and T.H. Jeng; critical revision of the manuscript for important intellectual content: D.W. Perng. All the authors discussed the results and approved the final version of the manuscript.

None declared.

The datasets generated or analyzed during this study are available from the corresponding author upon reasonable request.

Poster presentation, 2023 ATS conference, Washington, DC, May 21, 2023