The authors have recently identified an error in the data set used for the manuscript entitled “Machine Learning and Prediction of All-Cause Mortality in COPD” (CHEST 2020;158(3):952-964). The BODE index provided to the authors in the Visit 1 COPDGene dataset had been mistakenly calculated using a 6-min walk distance (6MWD) in feet rather than meters. This led to a systematic underestimation of the severity of the BODE score in COPDGene participants. This error specifically impacts all analyses in this manuscript utilizing the BODE index and e-BODE. The analyses the authors reported regarding “updated BODE” and the Age, Dyspnea, Obstruction (ADO) were not affected by the BODE error in the canonical COPDGene dataset as the authors calculated these indices directly from component measures. Fortunately, the key results and conclusions of the manuscript were not affected by this error.
The authors have recalculated the BODE index and e-BODE using meters for 6MWD and have repeated all analyses in the manuscript that involved either BODE or e-BODE. The predictive performance of the corrected BODE and e-BODE changed little compared to the initially reported analyses. The MLMP-COPD model continues to outperform all tested mortality prediction models including BODE, e-BODE, updated BODE, and ADO.
The corrected results from the analysis do alter the comparison of BODE quartiles and equal-sized MLMP-COPD groups (Fig 5, Table 3). The BODE group with the highest mortality (BODE score 7-10) now includes more than twice as many individuals (n = 466) compared to the original publication (n = 184), and the clinical characteristics between the highest mortality BODE group and same-sized highest predicted mortality MLMP-COPD group are different from the original report (Corrigendum Table 3). Percent emphysema and diabetes mellitus are now similar between groups; African-American race, 6MWD, BMI, FEV1/FVC ratio, and Pi10 (a measure of airway wall thickness), are now significantly different in the MLMP-COPD group. The MLMP-COPD group continues to have a greater proportion of deaths and a higher mean PA:A ratio.
Figure 5.
Kaplan-Meier analysis of subjects in a pooled COPDGene and ECLIPSE sample. A, Subjects were grouped by BODE score (0-2, 3-4, 5-6, 7-10). B, Subjects were stratified into four groups based on the machine learning mortality prediction (MLMP)-COPD model. Participants were ranked according to their calculated risk by the MLMP-COPD model and divided into four groups of mortality risk that were of equal size as those of the BODE groups. Dashed lines indicate the probability of survival at each time point, and the bands are 95% CIs. See Figure 1 and 3 legends for expansion of other abbreviations.
Table 3.
Characteristics of Subjects With the Highest Predicted Mortality
| Characteristic | BODE Group | Model Group | P Value |
|---|---|---|---|
| No. of subjects | 466 | 466 | |
| Age, mean (SD), y | 64.51 (8.25) | 68.03 (7.03) | < .001 |
| Race, African American, No. (%) | 81 (17.4) | 46 (9.9) | .001 |
| Sex, female, No. (%) | 211 (45.3) | 162 (34.8) | .001 |
| Total dead, No. (%) | 259 (55.6) | 290 (62.2) | .046 |
| Days followed up, median (IQR) | 1,809 (1,068-2,442) | 1,350 (914-2,230) | .002 |
| BODE, median (IQR) | 7.00 (7.00-8.00) | 7.00 (6.00-8.00) | < .001 |
| 6-Min walk distance, mean (SD), ft | 644.85 (281.57) | 692.98 (293.03) | .011 |
| FEV1 % predicted, median (IQR) | 28.60 (21.40-34.60) | 28.60 (21.60-36.40) | .548 |
| mMRC dyspnea score, No. (%) | < .001 | ||
| 0 | 0 (0.0) | 7 (1.5) | |
| 1 | 0 (0.0) | 22 (4.7) | |
| 2 | 27 (5.8) | 102 (21.9) | |
| 3 | 162 (34.8) | 173 (37.1) | |
| 4 | 277 (59.4) | 162 (34.8) | |
| FEV1/FVC ratio, median (IQR) | 0.35 (0.29-0.42) | 0.34 (0.28-0.39) | .001 |
| FEF25%-75%, median (IQR), L | 0.24 (0.19-0.33) | 0.24 (0.18-0.31) | .573 |
| Resting SaO2, median (IQR) | 94.00 (92.00-96.00) | 93.00 (91.00-96.00) | .002 |
| Exacerbation frequency, mean (SD), No./y | 1.25 (1.52) | 1.41 (1.47) | .101 |
| Pack-years cigarette smoking, median (IQR) | 49.00 (36.00-73.12) | 54.00 (40.00-80.00) | .002 |
| BMI, mean (SD), kg/m2 | 26.17 (6.94) | 25.39 (5.59) | .059 |
| Severe exacerbations, No. (%) | 190 (40.8) | 215 (46.1) | .113 |
| PA:A ratio, mean (SD) | 0.95 (0.16) | 0.99 (0.16) | < .001 |
| Pi10, mean (SD) | 3.91 (0.30) | 4.06 (0.37) | < .001 |
| Diabetes, No. (%) | 52 (11.2) | 66 (14.2) | .2 |
| % LAA < −950 HU, median (IQR) | 25.30 (14.01-37.02) | 24.62 (15.82-35.04) | .979 |
The BODE group with the highest predicted mortality had BODE score 7 to 10 (see Materials and Methods). See Table 1 and 2 legends for expansion of abbreviations.
Below, the authors detail the corrections made to the manuscript in each section. Critical changes are highlighted in red.
The online version of the article has been corrected. The authors regret this error.
Abstract (page 952)
The Results subsection should read “The group with the highest BODE scores (7-10) had 56% mortality, whereas the highest mortality group defined by the MLMP-COPD model had 62% mortality (P = .046).”
Subgroup Analyses (right column, page 957)
The text should read, “Characteristics of the sickest group defined by BODE (score 7-10) and our model are shown in Table 3, and the less severe groups in e-Tables 11, 12, and 13. The sickest MLMP-COPD group had a 11.9% higher relative and 6.6% absolute mortality than did the sickest BODE group (BODE 7-10; 56% vs 62% absolute mortality; P = .046), with shorter median follow-up (1,809 vs 1,350 days; P = .002). The sickest MLMP-COPD group was also older and had fewer female subjects, fewer African-Americans, higher PA:A ratio, higher Pi10, more pack-years of cigarette smoking, and lower resting arterial oxygen saturation. However, both groups had similar FEV1 % predicted, but the sickest MLMP-COPD group had lower modified Medical Research Council dyspnea scores, higher 6MWD, and lower BODE scores (P < .001).”
e-Table 11.
Characteristics of BODE group (score 0-2) compared to equally-sized MLMP-COPD mortality risk group
| BODE group | Model group | p-value | |
|---|---|---|---|
| No. | 1589 | 1589 | |
| Age, mean (SD), y | 63.44 (8.57) | 60.31 (8.11) | <0.001 |
| Race, African American, No. (%) | 166 (10.4) | 233 (14.7) | <0.001 |
| Sex, female, No. (%) | 603 (37.9) | 696 (43.8) | 0.001 |
| Total dead, No. (%) | 216 (13.6) | 163 (10.3) | 0.004 |
| Days followed up, median (IQR) | 2530.00 [2189.00, 2895.00] | 2530.00 [2199.00, 2834.00] | 0.885 |
| BODE, median (IQR) | 1.00 [0.00, 2.00] | 1.00 [0.00, 2.00] | <0.001 |
| 6-minute walk distance, mean (SD), ft | 1458.03 (273.13) | 1471.93 (289.34) | 0.164 |
| FEV1 % predicted, median (IQR) | 65.00 [55.20, 71.60] | 63.90 [53.80, 71.50] | 0.064 |
| mMRC dyspnea score, No. (%) | <0.001 | ||
| 0 | 623 (39.2) | 536 (33.7) | |
| 1 | 608 (38.3) | 430 (27.1) | |
| 2 | 286 (18.0) | 311 (19.6) | |
| 3 | 72 (4.5) | 229 (14.4) | |
| 4 | 0 (0.0) | 83 (5.2) | |
| FEV1/FVC ratio, median (IQR) | 0.57 [0.48, 0.63] | 0.58 [0.49, 0.64] | 0.002 |
| FEF25%-75%, median (IQR), L | 0.68 [0.47, 0.95] | 0.69 [0.48, 0.97] | 0.201 |
| Resting SaO2, median (IQR) | 96.00 [94.00, 97.00] | 96.00 [95.00, 97.00] | 0.003 |
| Exacerbation frequency, mean (SD), No./yr | 0.41 (0.84) | 0.40 (0.85) | 0.866 |
| Pack-years cigarette smoking, median (IQR) | 44.00 [32.20, 60.00] | 42.00 [31.00, 57.90] | 0.002 |
| BMI, mean (SD), kg/mˆ2 | 27.75 (5.01) | 28.60 (5.91) | <0.001 |
| Severe Exacerbations, No. (%) | 156 (9.8) | 143 (9.0) | 0.466 |
| PA:A ratio, mean (SD) | 0.88 (0.13) | 0.86 (0.12) | 0.009 |
| Pi10, mean (SD) | 3.90 (0.37) | 3.84 (0.33) | <0.001 |
| Diabetes, No. (%) | 149 (9.4) | 140 (8.8) | 0.622 |
| % LAA < -950 HU, median (IQR) | 6.24 [2.21, 14.67] | 5.02 [1.78, 13.28] | 0.001 |
e-Table 12.
Characteristics of BODE group (score 3-4) compared to equally-sized MLMP-COPD mortality risk group
| BODE group | Model group | p-value | |
|---|---|---|---|
| No. | 1066 | 1066 | |
| Age, mean (SD), y | 63.38 (8.26) | 64.58 (8.10) | 0.001 |
| Race, African American, No. (%) | 147 (13.8) | 126 (11.8) | 0.195 |
| Sex, female, No. (%) | 448 (42.0) | 426 (40.0) | 0.355 |
| Total dead, No. (%) | 275 (25.8) | 262 (24.6) | 0.549 |
| Days followed up, median (IQR) | 2319.50 [1736.00, 2798.00] | 2346.00 [1887.25, 2834.00] | 0.115 |
| BODE, median (IQR) | 3.00 [3.00, 4.00] | 3.00 [2.00, 4.00] | 0.01 |
| 6-minute walk distance, mean (SD), ft | 1214.41 (289.78) | 1178.44 (279.92) | 0.004 |
| FEV1 % predicted, median (IQR) | 47.25 [38.80, 57.70] | 47.50 [38.30, 58.88] | 0.632 |
| mMRC dyspnea score, No. (%) | <0.001 | ||
| 0 | 75 (7.0) | 131 (12.3) | |
| 1 | 161 (15.1) | 242 (22.7) | |
| 2 | 425 (39.9) | 284 (26.6) | |
| 3 | 329 (30.9) | 288 (27.0) | |
| 4 | 76 (7.1) | 121 (11.4) | |
| FEV1/FVC ratio, median (IQR) | 0.47 [0.39, 0.56] | 0.46 [0.39, 0.55] | 0.443 |
| FEF25%-75%, median (IQR), L | 0.42 [0.31, 0.62] | 0.41 [0.31, 0.59] | 0.619 |
| Resting SaO2, median (IQR) | 95.00 [93.00, 97.00] | 95.00 [94.00, 97.00] | 0.589 |
| Exacerbation frequency, mean (SD), No./yr | 0.80 (1.12) | 0.78 (1.16) | 0.635 |
| Pack-years cigarette smoking, median (IQR) | 46.30 [35.08, 64.33] | 47.40 [37.50, 65.75] | 0.112 |
| BMI, mean (SD), kg/mˆ2 | 27.96 (6.17) | 27.74 (5.94) | 0.41 |
| Severe Exacerbations, No. (%) | 236 (22.1) | 216 (20.3) | 0.314 |
| PA:A ratio, mean (SD) | 0.91 (0.14) | 0.90 (0.14) | 0.146 |
| Pi10, mean (SD) | 3.99 (0.38) | 3.98 (0.36) | 0.333 |
| Diabetes, No. (%) | 144 (13.5) | 131 (12.3) | 0.438 |
| % LAA < -950 HU, median (IQR) | 12.48 [4.98, 22.48] | 13.16 [5.34, 23.07] | 0.324 |
e-Table 13.
Characteristics of BODE group (score 5-6) compared to equally-sized MLMP-COPD mortality risk group
| BODE group | Model group | p-value | |
|---|---|---|---|
| No. | 779 | 779 | |
| Age, mean (SD), y | 63.44 (8.04) | 66.08 (7.39) | <0.001 |
| Race, African American, No. (%) | 91 (11.7) | 80 (10.3) | 0.418 |
| Sex, female, No. (%) | 321 (41.2) | 299 (38.4) | 0.277 |
| Total dead, No. (%) | 286 (36.7) | 321 (41.2) | 0.077 |
| Days followed up, median (IQR) | 2199.00 [1321.00, 2664.50] | 2156.00 [1188.50, 2708.50] | 0.614 |
| BODE, median (IQR) | 5.00 [5.00, 6.00] | 5.00 [4.00, 6.00] | <0.001 |
| 6-minute walk distance, mean (SD), ft | 990.52 (297.80) | 982.59 (292.66) | 0.596 |
| FEV1 % predicted, median (IQR) | 34.90 [27.80, 45.05] | 37.10 [28.20, 47.10] | 0.06 |
| mMRC dyspnea score, No. (%) | <0.001 | ||
| 0 | 5 (0.6) | 29 (3.7) | |
| 1 | 22 (2.8) | 97 (12.5) | |
| 2 | 175 (22.5) | 216 (27.7) | |
| 3 | 406 (52.1) | 279 (35.8) | |
| 4 | 171 (22.0) | 158 (20.3) | |
| FEV1/FVC ratio, median (IQR) | 0.38 [0.31, 0.47] | 0.39 [0.32, 0.47] | 0.761 |
| FEF25%-75%, median (IQR), L | 0.31 [0.22, 0.43] | 0.30 [0.22, 0.42] | 0.583 |
| Resting SaO2, median (IQR) | 95.00 [92.00, 96.00] | 94.00 [92.00, 96.00] | 0.047 |
| Exacerbation frequency, mean (SD), No./yr | 1.04 (1.31) | 0.99 (1.25) | 0.405 |
| Pack-years cigarette smoking, median (IQR) | 49.50 [38.00, 70.00] | 49.10 [39.00, 70.00] | 0.884 |
| BMI, mean (SD), kg/mˆ2 | 27.59 (6.62) | 26.60 (5.77) | 0.002 |
| Severe Exacerbations, No. (%) | 239 (30.7) | 247 (31.7) | 0.702 |
| PA:A ratio, mean (SD) | 0.93 (0.15) | 0.94 (0.14) | 0.415 |
| Pi10, mean (SD) | 3.97 (0.36) | 4.02 (0.37) | 0.003 |
| Diabetes, No. (%) | 89 (11.4) | 97 (12.5) | 0.584 |
| % LAA < -950 HU, median (IQR) | 18.57 [8.69, 30.11] | 18.91 [10.83, 29.69] | 0.149 |
Discussion (right column, page 961)
The text should read, “When stratifying subjects by using the MLMP-COPD model, the sickest group of the same-sized BODE group (score 7-10) had 11.9% higher mortality. Older age, lower resting arterial oxygen saturation, higher PA:A ratio, higher Pi10, and increased pack-years of smoking added prognostic information for these subjects. Thus, it is important to think beyond the BODE variables when trying to identify patients with COPD at the highest risk of death.”
Below, are the updated Figure 5, Table 3, and e-Table 11, e-Table 12, e-Table 13.
Supplementary Data
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.