RE-EVALUATION OF COMBINATION THERAPY IN CHRONIC OBSTRUCTIVE PULMONARY DISEASE (COPD)

Abstract

BACKGROUND:

Clinical trials of COPD pharmacotherapy typically involve aging populations with moderate-to-severe COPD, but the latter is often diagnosed by spirometric criteria that are not age-appropriate across the continuum of lung function. We have therefore re-evaluated the clinical effect of combination therapy (salmeterol plus fluticasone) in moderate-to-severe COPD, using more age-appropriate spirometric criteria from the Global Lung Function Initiative (GLI) and trial data from Towards a Revolution in COPD Health (TORCH).

METHODS:

Of the 6112 TORCH participants, 5688 (93.1%) had GLI-based moderate-to-severe COPD (mean age 64.8 years). The primary outcome was all-cause mortality and the primary comparison was combination therapy vs. placebo. Secondary outcomes included COPD and cardiovascular (CV) mortality and pneumonia. A modified intention-to-treat analysis evaluated differences in time-to-event over a three-year period, using Cox proportional hazards models with statistical significance at p <0.010 (acknowledging repeated significance testing).

RESULTS:

Relative to placebo, combination therapy yielded a statistically non-significant reduction in all-cause mortality—adjusted hazard ratio [adjHR] 0.78 (95% confidence interval [CI]: 0.64, 0.95), p=0.012. Relative to placebo, combination therapy also yielded statistically non-significant reductions in COPD and CV mortality—adjHR 0.75 (95% CI: 0.55, 1.02), p=0.068 and adjHR 0.76 (95% CI: 0.53, 1.09), p=0.135, respectively. In contrast, combination therapy yielded a statistically significant increased risk of pneumonia, relative to placebo— adjHR 1.80 (95% CI: 1.46, 2.21), p<0.001.

CONCLUSION:

In GLI-based moderate-to-severe COPD, combination therapy yields a statistically significant increased risk of pneumonia but the reductions in mortality are not statistically significant, although could potentially be clinically meaningful.

INTRODUCTION

COPD is commonly diagnosed by spirometric criteria from the Global initiative for chronic Obstructive Lung Disease (GOLD)^1,2 but these spirometric criteria have age-related limitations across the continuum of lung function.^3–11 Specifically, in middle-aged and older persons, the GOLD threshold of 0.70 for the ratio of forced expiratory volume in 1-second (FEV₁) to forced vital capacity (FVC) frequently misclassifies normal-for-age spirometry, restrictive-pattern, and airflow-obstruction (COPD).^3–11 Moreover, the GOLD-based use of FEV₁ percent predicted (%Pred) for establishing the severity of airflow-obstruction (COPD) assumes incorrectly that a given value is equivalently low or high across the adult lifespan.^3,4,7,8,11

Populations are rapidly aging worldwide and COPD predominantly affects middle-aged and older persons.^10,12,13 Concerns are therefore raised regarding the validity of clinical trials of COPD pharmacotherapy, as these often enroll middle-aged and older persons with GOLD-based moderate-to-severe airflow-obstruction (COPD).^14,15 For example, the clinical trial of Towards a Revolution in COPD Health (TORCH) compared combination therapy (salmeterol plus fluticasone) with placebo in participants having a mean age of 65 years and GOLD-based moderate-to-severe airflow-obstruction (COPD)—FEV₁/FVC≤0.70, FEV₁ <60%Pred.¹⁴

To establish more age-appropriate definitions of normal spirometry and spirometric impairments, including airflow-obstruction (COPD) and restrictive-pattern, a Z-score approach has been proposed by the Global Lung Function Initiative (GLI).³ The latter uses the Lambda-Mu-Sigma (LMS)⁴ method to calculate spirometric Z-scores, accounting for age-related changes in lung function and variability in spirometric performance. In clinical practice, Z-scores are routinely reported in bone mineral density testing and LMS is widely applied to growth charts.^4,16

Prior work¹⁵ has shown that GLI-based reclassification of the TORCH trial (N=6112) yields a substantially higher frequency of severe COPD (GLI-based 89.6% vs. GOLD-based 73.8%) and misclassification of GLI-based restrictive-pattern as GOLD-based moderate-to-severe COPD (6.9%). Notably, GLI-based severe COPD establishes an especially advanced stage of disease, as defined by clinical, physiological, and anatomical features,^{10,15,17–19} and GLI-based restrictive-pattern does not have an obstructive phenotype, since it is not associated with air trapping, hyperinflation, or emphysema.^10,17 In addition, GLI-based restrictive-pattern increases the risk of cardiovascular (CV) death and inappropriate use of COPD pharmacotherapy can lead to adverse CV outcomes and pneumonia.^20–22

In the current study, using Cox proportional hazards models and a sample limited to TORCH participants with GLI-based moderate-to-severe COPD, we have re-evaluated the effect of combination therapy on health outcomes, relative to placebo and over three years of follow-up. In supplemental analyses, as a basis for comparison, we have also applied the same Cox proportional hazards models to the original sample of TORCH participants with GOLD-based moderate-to-severe COPD. We posit that, since TORCH participants had a mean age of 65 years, an age-appropriate spirometric diagnosis of moderate-to-severe COPD should be a fundamental step when evaluating the effects of combination therapy on the primary TORCH outcome of all-cause mortality and secondary TORCH outcomes of COPD and CV mortality, and pneumonia.

MATERIALS AND METHODS

Data from the TORCH trial was accessed at clinicalstudydatarequest.com. Because the TORCH dataset was already de-identified and participants were not contacted, the current study was granted exemption from participant consent and ethical approval by the institutional review board of Yale University.

TORCH Trial

The TORCH trial enrolled 6112 participants with mean age of 65 years and GOLD-based moderate-to-severe COPD, diagnosed by pre-bronchodilator FEV₁/FVC ≤0.70 and FEV₁ <60%Pred.¹⁴ The trial included four treatment groups: 1) 50 μg of salmeterol (long acting beta-2 agonist [LABA] bronchodilator) plus 500 μg of fluticasone priopionate (corticosteroid), i.e., combination therapy; 2) salmeterol alone (50 μg); 3) fluticasone alone (500 μg); and 4) placebo — administered as a single inhaler twice daily over a three-year period.

The TORCH trial protocol and analytical plan can be accessed at ClinicalTrials.gov, study number NCT00268216, or at the GlaxoSmithKline (GSK) Clinical Study Register, through the URL: https://www.clinicalstudydatarequest.com.

Baseline Characteristics

As shown in Table 1, the baseline characteristics included age, height, sex, race/ethnicity (fixed set of options), smoking history, and respiratory phenotypes, namely dyspnea (Modified Medical Research Council), prior COPD exacerbation, bronchodilator reversibility, and severity of spirometric airflow-obstruction.

Table 1.

Baseline characteristics of TORCH participants with GLI-based moderate-to-severe COPD (N=5688)

Characteristic	All (N=5688)	Placebo (N=1413)	Salmeterol (N=1418)	Fluticasone (N=1427)	Combination * (N=1430)	P value†
		Mean ± SD or No. (Column %)
Age, years	64.8 ±8.3	64.8 ±8.2	64.8 ±8.2	64.8 ±8.5	64.8 ±8.3	0.997
Aged ≥60 years	4210 (74.0)	1040 (73.6)	1068 (75.3)	1051 (73.6)	1051 (73.5)	0.643
Height, meters	1.7 ±0.1	1.7 ±0.1	1.7 ±0.1	1.7 ±0.1	1.7 ±0.1	0.417
Male sex	4303 (75.6)	1069 (75.6)	1087 (76.7)	1078 (75.5)	1069 (74.8)	0.703
BMI	25.2 ±5.1	25.3 ±5.1	25.2 ±5.1	25.1 ±5.0	25.2 ±5.3	0.905
Race/Ethnicity
White‡	4622 (81.3)	1151 (81.5)	1154 (81.4)	1154 (80.9)	1163 (81.3)	0.999
American Hispanic§	179 (3.1)	46 (3.3)	42 (3.0)	45 (3.1)	46 (3.2)
African-American¶	82 (1.4)	21 (1.5)	18 (1.3)	21 (1.5)	22 (1.5)
Asian **	756 (13.3)	185 (13.1)	190 (13.4)	194 (13.6)	187 (13.1)
Other††	49 (0.9)	10 (0.7)	14 (1.0)	13 (0.9)	12 (0.8)
Smoking history
Current smokers	2465 (43.3)	613 (43.4)	610 (43.0)	621 (43.5)	621 (43.4)	0.994
Former smokers	3223 (56.7)	800 (56.6)	808 (57.0)	806 (56.5)	809 (56.6)
Pack-years‡‡	48.5 ±27.4	48.7 ±27.0	49.3 ±27.8	49.1 ±28.4	47.1 ±26.2	0.217
Respiratory Phenotypes
MMRC dyspnea grade ≥ 2§§	2855 (50.2)	700 (49.6)	707 (49.9)	734 (51.5)	714 (49.9)	0.719
Prior COPD exacerbation¶¶	0.2 ±0.6	0.2 ±0.7	0.2 ±0.6	0.3 ±0.6	0.2 ±0.6	0.610
Spirometry
Pre-BD FEVi/FVC ratio	0.47 ±0.10	0.47 ±0.10	0.47 ±0.10	0.47 ±0.10	0.47 ±0.10	0.792
Pre-BD FEVi in liters	1.10 ±0.39	1.10 ±0.39	1.09 ±0.39	1.10 ±0.39	1.10 ±0.40	0.913
Pre-BD FVC in liters	2.35 ±0.76	2.37 ±0.75	2.32 ±0.73	2.37 ±0.79	2.35 ±0.76	0.575
BD reversibility ***	2302 (40.5)	561 (39.7)	589 (41.5)	557 (39.1)	595 (41.6)	0.398
GLI airflow-obstruction†††
Moderate-to-severe	5688 (100)	1413 (100)	1418 (100)	1427 (100)	1430 (100)	------
Moderate	212 (3.7)	47 (3.3)	51 (3.6)	61 (4.3)	53 (3.7)	0.595
Severe	5476 (96.3)	1366 (96.7)	1367 (96.4)	1366 (95.7)	1377 (96.3)

Abbreviations: BD, bronchodilator; BMI, body mass index; COPD, chronic obstructive pulmonary disease; GLI, Global Lung Function Initiative; MMRC, Modified Medical Research Council; TORCH Study, Towards a Revolution in COPD Health Study.

^*The combination group includes salmeterol plus fluticasone.

^†Comparisons were across treatment and placebo groups. Pearson chi-square tests were used for categorical outcomes and Kruskal-Wallis tests were used for continuous outcomes.

^‡Defined by TORCH as having origins in the original peoples of Europe, Middle East, Western Russia, Afghanistan, or white racial groups of Africa.

^§Defined by TORCH as Hispanics of North, Central, or South American origin.

^¶Defined by TORCH as having origins in any of the black racial groups of Africa.

^**Defined by TORCH as having origins in the original peoples of the Indian subcontinent, Far East, Southeast Asia, or Pacific Islands.

^††Defined by TORCH as other racial group that is not represented above, or whose predominant origin cannot be determined.

^‡‡Calculated as average pack-years for the combined group of current and former smokers.

^§§Grade ranges from 0–4. Missing data in 5 participants.

^¶¶Number of self-reported exacerbations of COPD that required hospitalization in the year prior to study entry.

^***Change in FEV₁, calculated as [(Post BD – pre-BD) / pre-BD FEV₁] x 100%. BD reversibility was defined by ATS-ERS criteria, specifically as an increase in the post-BD FEV₁ of >12% and ≥200 mL. Missing data in 1 participant.

^†††In the TORCH trial, airflow-obstruction confirmed a diagnosis of COPD. Using GLI reference equations, a combined diagnosis of moderate-to-severe airflow-obstruction (COPD) was established by Z-scores for FEV₁/FVC and FEV₁, both < −1.64. The distinction between moderate and severe airflow-obstruction (COPD) was further stratified by Z-scores for FEV₁ of < −1.64 but ≥ −2.55 as moderate disease and of < −2.55 for severe disease. The Z-score of −1.64 defined the 5^th percentile of distribution (lower limit of normal), while the Z-score of −2.55 defined the 0.5^th percentile of distribution.

Spirometry

The TORCH trial established spirometry-confirmed COPD based on pre-bronchodilator values. The use of pre-bronchodilator values has several distinct advantages: older persons have a reduced capacity to perform multiple pre- and post-bronchodilator FVC maneuvers and may experience an adverse response to a bronchodilator; post-bronchodilator values have limited relevance in distinguishing COPD from asthma and low reproducibility over time; and diagnostic thresholds for spirometric interpretation are based on reference populations that only recorded the equivalent of pre-bronchodilator values.^3,4,23–27

In contrast to the TORCH trial, however, we established a diagnosis of moderate-to-severe COPD based on pre-bonchodilator spirometric Z-scores < −1.64 for FEV₁/FVC and FEV₁, respectively. A Z-score of −1.64 defined the 5^th percentile of distribution, otherwise known as the lower limit of normal (LLN). To calculate spirometric Z-scores for FEV₁/FVC and FEV₁, we used TORCH-reported measured values from clinicalstudydatarequest.com and predicted values from sex- and ethnicity-specific GLI reference equations.³ These equations applied the LMS method,⁴ incorporating: the mean (Mu), representing how spirometric measures change based on predictor variables (age and height); the coefficient-of-variation (Sigma), representing the spread of reference values (variability in spirometric performance); and skewness (Lambda), representing departure from normality. Spirometric Z-scores have a strong mathematical, physiological, and clinical rationale, previously validated across the continuum of lung function in multiple cohorts of middle-aged and older persons.^{3–11,15,17–20}

Also, in contrast to the TORCH trial,¹⁴ we defined bronchodilator reversibility based on criteria from the American Thoracic Society and European Respiratory Society (ATS-ERS),²⁸ as an increase in post-bronchodilator FEV₁ of >12% and ≥200 mL. TORCH defined bronchodilator reversibility as an increase in post-bronchodilator FEV₁ of > 10% of the predicted value for the patient.¹⁴

Treatment Outcomes and Comparisons

Outcomes were centrally adjudicated by TORCH investigators, blinded to treatment assignment.¹⁴ As in the TORCH trial, our primary outcome was all-cause mortality and our secondary outcomes included cause-specific mortality (COPD and CV) and the occurrence of first pneumonia (regardless of severity or hospitalization). Also, as in the TORCH trial, our primary treatment comparison was combination therapy vs. placebo and our secondary treatment comparisons included combination therapy vs. salmeterol alone, combination therapy vs. fluticasone alone, salmeterol alone vs. placebo, and fluticasone alone vs. placebo. Treatment outcomes and comparisons were evaluated over a three-year intervention period.

Statistical Analysis

The primary analysis of the current study evaluated GLI-based moderate-to-severe COPD. Specifically, baseline characteristics were first summarized as means and standard deviations (SD), or as counts and percentages, with results stratified by treatment group. Next, using Cox proportional hazards models, comparisons yielding hazard ratios were made between treatment groups, over a three-year intervention period. In the Cox proportional hazards models, the outcomes were differences in time to death (all-cause, COPD, and CV) and first pneumonia. The covariates included age, sex, BMI, race/ethnicity, current smoker, pack-years, dyspnea, COPD exacerbation with hospitalization (in the year prior to study entry), and bronchodilator reversibility. In a supplemental analysis, a Fine and Gray competing risk of death analysis of the time to first pneumonia was also performed. This addressed the concern of whether death diminished the association between pharmacotherapy and the outcome of first pneumonia

As a basis for comparison, additional supplemental analyses evaluated the baseline characteristics of TORCH participants with GOLD-based moderate-to-severe COPD (as defined in the TORCH trial) and GLI-based restrictive-pattern (misclassified in the TORCH trial as GOLD-based moderate-to-severe COPD). Furthermore, the same Cox proportional hazards models and covariates, as used in our primary analyses of all-cause and cause-specific mortality, were applied to a sample of TORCH participants with GOLD-based moderate-to-severe COPD. Replicating the analytical plan of the original TORCH trial was otherwise difficult, given its design as a randomized controlled trial and adjustments for two interim analyses.¹⁴

All analyses of primary and secondary outcomes were performed according to a modified intention-to-treat principle. Specifically, participants retained in the analytical samples were evaluated as originally randomized, regardless of post-randomization adherence and attrition. SAS/STAT 13.1 software was used in all analyses. Results were interpreted to have statistical significance at p values <0.010, acknowledging repeated significance testing.

RESULTS

Of the original 6112 TORCH participants with GOLD-based moderate-to-severe COPD, 5688 (93.1%) had GLI-based moderate-to-severe COPD, which defined our primary analytical sample. We thus excluded 424 TORCH participants, of whom 420 had GLI-based restrictive-pattern, 3 had GLI-based normal-for age spirometry, and 1 had GLI-based mild COPD.

Table 1 reports the baseline characteristics of participants with GLI-based moderate-to-severe COPD. Overall, participants were largely aged ≥60 years, male, white, heavy smokers, and had moderate-to-severe dyspnea. In the year prior to study entry, COPD exacerbations with hospitalization was uncommon. Bronchodilator reversibility was established in 40.5% of participants, and almost all participants had GLI-based severe airflow-obstruction (96.3%). Otherwise, there were no differences in baseline characteristics across treatment groups.

As a basis for (nominal) comparisons, supplemental Table A1 (Appendix A) shows that baseline characteristics were similar in participants with GOLD- and GLI-based moderate-to-severe COPD. However, compared with GOLD- and GLI-based moderate-to-severe COPD, participants with GLI-based restrictive-pattern were older, had a higher BMI, were more frequently white, had a lower representation of current smokers, and had a higher FEV₁/FVC and FEV₁ but a lower FVC and frequency of bronchodilator reversibility.

Table 2 reports three-year all-cause mortality for GLI-based moderate-to-severe COPD. In the primary treatment comparison, combination therapy yielded an adjusted 22% reduction in all-cause mortality, relative to placebo—adjusted hazard ratio [adjHR] 0.78 (95% confidence interval [95%CI]: 0.64, 0.95), p=0.012. In secondary treatment comparisons, combination therapy yielded an adjusted 22% reduction in all-cause mortality, relative to fluticasone alone—adjHR 0.78 (95%CI: 0.64, 0.94), p=0.011. The other secondary treatment comparisons yielded substantially higher p values, ranging from 0.055 to 0.995.

Table 2.

Three-year analysis of all-cause mortality, as the primary outcome, in GLI-based moderate-to-severe COPD

All-Cause Mortality
Treatment Group *	No. (%) of Deaths†		Treatment Comparison‡	Hazard Ratio (95%CI)	P Value
Placebo (N=1413)	217 (15.4)		Unadjusted
Salmeterol (N=1418)	190 (13.4)		Combination vs. placebo	0.80 (0.66, 0.98)	0.029
Fluticasone (N=1427)	228 (16.0)		Combination vs. salmeterol	0.93 (0.76, 1.14)	0.495
Combination (N=1430)	179 (12.5)		Combination vs. fluticasone	0.77 (0.64, 0.94)	0.010
			Salmeterol vs. placebo	0.86 (0.71, 1.05)	0.135
			Fluticasone vs. placebo	1.04 (0.86, 1.25)	0.687
			Adjusted§
			Combination vs. placebo	0.78 (0.64, 0.95)	0.012
			Combination vs. salmeterol	0.94 (0.77, 1.15)	0.543
			Combination vs. fluticasone	0.78 (0.64, 0.94)	0.011
			Salmeterol vs. placebo	0.83 (0.68, 1.00)	0.055
			Fluticasone vs. placebo	1.00 (0.83, 1.21)	0.995

Abbreviations: CI, confidence interval; COPD, chronic obstructive pulmonary disease; GLI, Global Lung Function Initiative.

^*Combination group includes salmeterol plus fluticasone.

^†Row percent.

^‡The primary treatment comparison is in bold, i.e., combination therapy (salmeterol plus fluticasone) vs. placebo. The other treatment comparisons were pre-specified as secondary in the TORCH trial.

^§Adjusted for age, sex, BMI, race/ethnicity, current smoker, pack-years, dyspnea, prior COPD exacerbation, and bronchodilator reversibility.

As a basis for comparison, supplemental Table A2 (Appendix A) shows that combination therapy in GOLD-based moderate-to-severe COPD yielded an adjusted 21% reduction in all-cause mortality, relative to placebo—adjHR 0.79 (95%CI: 0.66, 0.96), p=0.018, which is similar to the GLI-based values in Table 2. The same pattern of results was shown for secondary treatment comparisons.

Table 3 reports three-year COPD mortality for GLI-based moderate-to-severe COPD. In the primary treatment comparison, combination therapy yielded an adjusted 25% reduction in COPD mortality, relative to placebo—adjHR 0.75 (95%CI: 0.55, 1.02), p=0.068. In the secondary treatment comparisons, combination therapy yielded an adjusted 32% reduction in COPD mortality, relative to fluticasone alone—adjHR of 0.68 (95%CI: 0.50, 0.93), p=0.015. The other secondary treatment comparisons yielded substantially higher p values, ranging from 0.189 to 0.555.

Table 3.

Three-year analysis of COPD mortality, as a secondary outcome, in GLI-based moderate-to-severe COPD

COPD Mortality *
Treatment Group†	No. (%) of COPD Deaths‡		Treatment Comparison§	Hazard Ratio (95%CI)	P Value
Placebo (N=1413)	87 (6.2)		Unadjusted
Salmeterol (N=1418)	87 (6.1)		Combination vs. placebo	0.79 (0.58, 1.09)	0.149
Fluticasone (N=1427)	100 (7.0)		Combination vs. salmeterol	0.81 (0.59, 1.10)	0.179
Combination (N=1430)	71 (5.0)		Combination vs. fluticasone	0.70 (0.51, 0.95)	0.021
			Salmeterol vs. placebo	0.98 (0.73, 1.32)	0.917
			Fluticasone vs. placebo	1.14 (0.85, 1.51)	0.382
			Adjusted¶
			Combination vs. placebo	0.75 (0.55, 1.02)	0.068
			Combination vs. salmeterol	0.81 (0.59, 1.11)	0.189
			Combination vs. fluticasone	0.68 (0.50, 0.93)	0.015
			Salmeterol vs. placebo	0.92 (0.68, 1.24)	0.587
			Fluticasone vs. placebo	1.09 (0.82, 1.46)	0.555

Abbreviations: CI, confidence interval; COPD, chronic obstructive pulmonary disease; GLI, Global Lung Function Initiative.

^*Centrally adjudicated and blinded to treatment assignment, using medical records and other data, as available.

^†Combination group includes salmeterol plus fluticasone.

^‡Row percent.

^§The primary treatment comparison is in bold, i.e., combination therapy (salmeterol plus fluticasone) vs. placebo. The other treatment comparisons were prespecified as secondary in the TORCH trial.

^¶Adjusted for age, sex, BMI, race/ethnicity, current smoker, pack-years, dyspnea, prior COPD exacerbation, and bronchodilator reversibility.

As a basis for comparison, supplemental Table A3 (Appendix A) shows that combination therapy in GOLD-based moderate-to-severe COPD yielded an adjusted 27% reduction in COPD mortality, relative to placebo—adjHR 0.73 (95%CI: 0.54, 1.00), p=0.047, which is similar to the GLI-based values in Table 3. The same pattern of results was shown for secondary treatment comparisons.

Table 4 reports three-year CV mortality for GLI-based moderate-to-severe COPD. In the primary treatment comparison, combination therapy yielded an adjusted 24% reduction in CV mortality, relative to placebo—adjHR 0.76 (95%CI: 0.53, 1.09), p=0.135. In the secondary treatment comparisons, salmeterol alone yielded an adjusted 41% reduction in CV mortality, relative to placebo—adjHR 0.59 (95%CI: 0.40, 0.88), p=0.001. The other secondary treatment comparisons yielded substantially higher p values, ranging from 0.234 to 0.759.

Table 4.

Three-year analysis of cardiovascular (CV) mortality, as a secondary outcome, in GLI-based moderate-to-severe COPD

CV Mortality *
Treatment Group†	No. (%) of C V Deaths‡		Treatment Comparison§	Hazard Ratio (95%CI)	P Value
Placebo (N=1413)	67 (4.7)		Unadjusted
Salmeterol (N=1418)	41 (2.9)		Combination vs. placebo	0.77 (0.54, 1.10)	0.157
Fluticasone (N=1427)	56 (3.9)		Combination vs. salmeterol	1.28 (0.85, 1.92)	0.237
Combination (N=1430)	53 (3.7)		Combination vs. fluticasone	0.93 (0.64, 1.36)	0.717
			Salmeterol vs. placebo	0.60 (0.41, 0.89)	0.011
			Fluticasone vs. placebo	0.83 (0.58, 1.18)	0.292
			Adjusted¶
			Combination vs. placebo	0.76 (0.53, 1.09)	0.135
			Combination vs. salmeterol	1.28 (0.85, 1.92)	0.236
			Combination vs. fluticasone	0.94 (0.65, 1.37)	0.759
			Salmeterol vs. placebo	0.59 (0.40, 0.88)	0.001
			Fluticasone vs. placebo	0.81 (0.57, 1.15)	0.234

Abbreviations: CI, confidence interval; COPD, chronic obstructive pulmonary disease; GLI, Global Lung Function Initiative.

^*Centrally adjudicated and blinded to treatment assignment, using medical records and other data, as available.

^†Combination group includes salmeterol plus fluticasone.

^‡Row percent.

^§The primary treatment comparison is in bold, i.e., combination therapy (salmeterol plus fluticasone) vs. placebo. The other treatment comparisons were prespecified as secondary in the TORCH trial.

^¶Adjusted for age, sex, BMI, race/ethnicity, current smoker, pack-years, dyspnea, prior COPD exacerbation, and bronchodilator reversibility.

As a basis for comparison, supplemental Table A4 (Appendix A) shows that combination therapy in GOLD-based moderate-to-severe COPD yielded an adjusted 19% reduction in CV mortality, relative to placebo—adjHR 0.81 (95%CI: 0.58, 1.15), p=0.241, which approximates the GLI-based values in Table 4. The same pattern of results was shown for secondary treatment comparisons.

Table 5 reports three-year risk of first pneumonia for GLI-based moderate-to-severe COPD. In the primary treatment comparison, combination therapy yielded an adjusted 80% increased risk of pneumonia, relative to placebo—adjHR 1.80 (95%CI: 1.46, 2.21), p<0.001. In the secondary treatment comparisons, combination therapy yielded an adjusted 57% increased risk of pneumonia, relative to salmeterol alone—adjHR 1.57 (95%CI: 1.29, 1.92), p<0.001, while fluticasone alone yielded an adjusted 65% increased risk of pneumonia, relative to placebo—adjHR 1.65 (95%CI: 1.34, 2.04), p<0.001. Otherwise, salmeterol alone yielded only a 14% increased risk of pneumonia, relative to placebo—adjHR 1.14 (95%CI: 0.91, 1.43), p=0.255. These results remained similar after accounting for the competing risk of death (see Appendix B, Table B1).

Table 5.

Three-year analysis of first pneumonia, as a secondary outcome, in GLI-based moderate-to-severe COPD

First Pneumonia *
Treatment Group†	No. (%) with First Pneumonia‡		Treatment Comparison§	Hazard Ratio (95%CI)	P Value
Placebo (N=1413)	138 (9.8)		Unadjusted
Salmeterol (N=1418)	162 (11.4)		Combination vs. placebo	1.80 (1.46, 2.21)	<0.001
Fluticasone (N=1427)	227 (15.9)		Combination vs. salmeterol	1.54 (1.26, 1.88)	<0.001
Combination (N=1430)	246 (17.2)		Combination vs. fluticasone	1.07 (0.89, 1.28)	0.470
			Salmeterol vs. placebo	1.17 (0.93, 1.46)	0.186
			Fluticasone vs. placebo	1.68 (1.36, 2.08)	<0.001
			Adjusted¶
			Combination vs. placebo	1.80 (1.46, 2.21)	<0.001
			Combination vs. salmeterol	1.57 (1.29, 1.92)	<0.001
			Combination vs. fluticasone	1.09 (0.91, 1.43)	0.361
			Salmeterol vs. placebo	1.14 (0.91, 1.43)	0.255
			Fluticasone vs. placebo	1.65 (1.34, 2.04)	<0.001

Abbreviations: CI, confidence interval; COPD, chronic obstructive pulmonary disease; GLI, Global Lung Function Initiative.

^*Centrally adjudicated and blinded to treatment assignment, using medical records and other data, as available. Includes a first pneumonia event (regardless of severity or hospitalization) over three years of follow-up.

^†Combination group includes salmeterol plus fluticasone.

^‡Row percent.

^§The primary treatment comparison is in bold, i.e., combination therapy (salmeterol plus fluticasone) vs. placebo. The other treatment comparisons were prespecified as secondary in the TORCH trial.

^¶Adjusted for age, sex, BMI, race/ethnicity, current smoker, pack-years, dyspnea, prior COPD exacerbation, and bronchodilator reversibility.

DISCUSSION

Given that populations worldwide are rapidly aging and COPD predominantly affects middle-aged and older persons,^10,12,13 an age-appropriate spirometric diagnosis should be a fundamental step when evaluating the clinical effects of COPD pharmacotherapy. Using data from the TORCH trial but a more age-appropriate, GLI-based spirometric diagnosis of moderate-to-severe COPD, we have therefore re-evaluated the clinical effect of combination therapy (salmeterol plus fluticasone) on the TORCH primary outcome of all-cause mortality.

Our results show that the risk reduction in all-cause mortality by combination therapy, relative to placebo, was not statistically significant. Specifically, the p value of 0.012 for the adjusted 22% risk reduction in death by the combination therapy did not meet the statistical significance criterion of p <0.010 (acknowledging repeated significance testing). Nonetheless, since p values alone provide limited information,²⁹ we posit that the 22% risk reduction in death may be clinically meaningful, given that our TORCH sample had a mean age of 64.8 years and had COPD which was predominantly severe by GLI criteria (96.3%). Older age is an especially strong predictor of death and GLI-based severe COPD is an especially advanced stage of disease, associated with 5-fold higher odds of moderate-to-severe dyspnea, 3-fold higher odds of emphysema (volumetric chest computed tomography [CT]), and 3-fold higher risk of COPD hospitalization, as compared with GLI-based moderate COPD.^{10,15,17–20} Moreover, we posit that our statistical power may have been limited by the GLI reclassification of the TORCH trial, as this decreased our analytical sample from 6112 to 5688 participants. Accordingly, future clinical trials of COPD pharmacotherapy should consider a larger sample of participants with GLI-based COPD, including greater representation of moderate COPD.

Regarding secondary treatment outcomes, our results show that risk reductions in COPD and CV mortality by combination therapy, relative to placebo, were not statistically significant. In particular, the adjusted 25% risk reduction in COPD mortality and the adjusted 24% risk reduction in CV mortality yielded p values of 0.068 and 0.135, respectively, which were higher than the statistical significance criterion of p <0.010. Nonetheless, for reasons already discussed, the noted risk reductions may be clinically meaningful.

Conversely, combination therapy yielded an adjusted, statistically significant 80% increased risk of pneumonia (p<0.001), relative to placebo. This effect was due to the fluticasone component of the combination therapy, since fluticasone alone yielded an adjusted, statistically significant 65% increased risk of pneumonia (p<0.001), whereas salmeterol alone did not have a substantive effect (p=0.255) — both comparisons were relative to placebo. As to underlying mechanisms, inhaled fluticasone may modify local immunological factors within the respiratory system which, in turn, increase the risk of pneumonia.³⁰ The adverse effect of incident pneumonia highlights the need for influenza and pneumococcal vaccinations when prescribing combination therapy, as well as the need to develop an alternative to inhaled corticosteroids as the anti-inflammatory component of combination therapy.

Regarding secondary treatment comparisons, our results show that salmeterol alone yielded an adjusted, statistically significant 41% risk reduction in CV mortality (p=0.001) but had no substantive effect on COPD mortality (8% risk reduction, p=0.587), relative to placebo. These results contrast with prior work, showing that initiation of a LABA increased the risk of adverse CV outcomes in COPD.²¹ However, whereas the prior study used health administrative data,²¹ we used spirometric criteria to establish COPD and excluded GLI-based restrictive-pattern. As discussed earlier, GLI-based restrictive-pattern does not have an obstructive phenotype and increases the risk of CV death.^10,17,20 As to mechanisms underlying the CV benefit of salmeterol in COPD, we hypothesize improved myocardial performance and/or pulmonary vascular resistance,^31,32 potentially due to improved work of breathing.³³

As a basis for comparison, the current study has undertaken a supplemental analysis of nominal differences in baseline characteristics across three spirometric groups. These included GLI-based moderate-to-severe COPD (our primary sample) and GLI-based restrictive-pattern (misclassified in the TORCH trial as GOLD-based moderate-to-severe COPD), which together comprised 99.9% of the original TORCH sample of GOLD-based moderate-to-severe COPD. Across these three spirometric groups, participants with GLI-based restrictive-pattern were older and had a higher BMI, lower representation of current smokers, normal-for-age FEV₁/FVC, and higher FEV₁. Otherwise, GOLD- and GLI-based moderate-to-severe COPD had similar baseline characteristics.

Also, as a basis for comparison, an additional supplemental analysis evaluated treatment effects on mortality in the original TORCH sample of GOLD-based moderate-to-severe COPD, using the same Cox proportional hazards models and covariates as in our primary analysis. Our results show that combination therapy, relative to placebo, yielded similar associations with all-cause, COPD, and CV mortality in GOLD- and GLI-based moderate-to-severe COPD. Hence, excluding GLI-based restrictive-pattern from the TORCH sample did not change treatment effects.

Given that GLI- and GOLD-based criteria for moderate-to-severe COPD yielded similar baseline characteristics and treatment effects, which spirometric criteria should be used in clinical trials of pharmacotherapy? Ultimately, the merits of spirometric criteria in clinical trials of COPD pharmacotherapy should be further defined by their applicability to general clinical practice. In patient populations, wherein multimorbidity and polypharmacy occur commonly and adverse health effects result from the inappropriate use of medications or delays in considering alternative diagnoses, a fundamental step in disease management is an age-appropriate evaluation.^{3–11,15,34–37} Thus, spirometric criteria in general clinical practice need to be age-appropriate across the continuum of lung function, since excluding disease (normal-for-age spirometry) is as important as establishing disease (spirometric restrictive-pattern or airflow-obstruction [COPD]).^11,37

Prior work has already shown that GLI-based criteria, but not GOLD-based criteria, have strong validation across the continuum of lung function.^{3–11,15,17–20} For example, using data from COPDGene (Genetic Epidemiology of COPD) and from a large sample of patients who had completed pulmonary function tests, the same GLI-calculated spirometric Z-score thresholds that were used in the current study have established: normal-for-age spirometry as having a normal respiratory phenotype; spirometric restrictive pattern as having a restrictive ventilatory defect (lower total lung capacity); and mild, moderate, and severe spirometric airflow-obstruction (COPD) as having graded associations with dyspnea, respiratory health related quality of life, exercise performance, CT-measured emphysema and gas trapping, hyperinflation (higher total lung capacity), and gas exchange (lower single-breath diffusing capacity).^9,10,17 In contrast, it is already well established that GOLD criteria misclassify normal-for-age spirometry, restrictive-pattern and airflow-obstruction (COPD).^3–11

Accordingly, future clinical trials of COPD pharmacotherapy should strongly consider the use of a more age-appropriate, GLI-based diagnosis of COPD for two major reasons. First, GLI-based criteria, but not GOLD-based criteria, have strong validation across the continuum of lung function and, thus, can be integrated into general clinical practice, especially when managing patients with complex multimorbidity.^{3–11,15,17–20} Second, the results of the current study did not show even a nominal, much less statistically significant difference in treatment effects for the GLI and GOLD study samples. Consequently, there is little to no rationale for the continued use of GOLD-based criteria in clinical trials of COPD pharmacotherapy.

In the current study, since the sample size for GLI-based restrictive-pattern was small (n=420 [6.9%]), the efficacy and safety of combination therapy could not be evaluated in this group of patients. At a prevalence of 6.9% in the TORCH trial, we estimate that misclassification of GLI-based restrictive-pattern as GOLD-based moderate-to-severe COPD may involve more than 165,000 Americans (0.069 * 2.4 million with GOLD-based severe COPD).³⁸ As discussed earlier, GLI-based restrictive-pattern does not have an obstructive phenotype and may increase the risk of CV death.^10,17,20 Future work should therefore evaluate the treatment effect of combination therapy in a larger study population of patients who have GLI-based restrictive-pattern misclassified as GOLD-based COPD.

We acknowledge that re-evaluating a clinical trial from 2007 (TORCH)¹⁴ may be viewed as a limitation. We note, however, that combination therapy remains a frequent option in COPD but its effect on clinical outcomes required further evaluation, since age-appropriate spirometric criteria were not used in the original TORCH report.¹⁴ Currently, age-appropriate spirometric criteria are a high priority, given that populations worldwide (and in clinical practice) are rapidly aging and that COPD continues to predominantly affect middle-aged and older persons.^10,12,13,35

Other limitations are further acknowledged. Foremost, the benefits of randomization may have been lost with GLI reclassification of the TORCH trial (unremedied confounding may have been introduced) and the two interim analyses of the TORCH trial could not be replicated.¹⁴ These limitations confer an exploratory nature to our study, which can only be definitively addressed by a new clinical trial with GLI-based enrollment criteria. In the interim, despite the stated limitations of the current study, we note that our GLI-reclassification of treatment groups yielded very similar baseline characteristics, that important covariates were incorporated into our respective models, and that our level of statistical significance was defined by p<0.010 (acknowledging repeated significance testing).³⁹ Other limitations specific to the TORCH trial were previously discussed in the original report and accompanying editorial.^14,40

Lastly, we acknowledge two other approaches for staging COPD severity. Using GLI-calculated FEV₁ Z-scores, a five-level staging of COPD severity has been previously proposed.⁴¹ We did not apply this stratification, because a threshold value at the LLN (Z-score of −1.64) was not included.⁴¹ We posit that the LLN strongly merits being used as a threshold value when staging FEV₁, since it is a standard criterion for establishing normal-for-age lung function.²⁸ The second approach is based on ATS-ERS criteria,²⁸ but the latter uses %Pred thresholds for FEV₁. As discussed earlier, the use of %Pred is not age-appropriate and may result in misclassification of COPD severity in aging populations (e.g., moderate and severe COPD by Z-scores may be staged as mild and moderate COPD by the ATS-ERS criteria, respectively).^7,8,42

CONCLUSION

Using data from the TORCH trial but a more age-appropriate spirometric diagnosis of moderate-to-severe COPD, our results suggest that combination therapy yields a statistically significant increased risk of pneumonia but the reductions in mortality are not statistically significant, although could potentially be clinically meaningful. Given that populations worldwide are rapidly aging and COPD predominantly affects middle-aged and older persons,^10,12,13 our results inform the rationale for implementing more age-appropriate spirometric criteria in future clinical trials of COPD pharmacotherapy.

Highlights:

• Re-classified the TORCH trial, using age-appropriate spirometric criteria from GLI

• Re-evaluated combination therapy in GLI-defined moderate-to-severe COPD

• Combination therapy yielded a statistically non-significant reduction in mortality

• Combination therapy yielded a statistically significant increased risk of pneumonia