Table 4– Sensitivity analysis: Cox regression models to predict 1-year and 2-year all-cause mortality in patients with stable chronic obstructive pulmonary disease, with missing 6-min walk test data imputed (Hotdeck imputation)

	Likelihood ratio Chi-squared	p-value	C statistic
1-year mortality
Univariate analysis
Pro-adrenomedullin	24.38	<0.001	0.715
BODE	33.55	<0.001	0.759
BOD^#	24.41	<0.001	0.714
Multivariate analysis
Pro-adrenomedullin plus:
BODE	53.33^¶	<0.001	0.829
BOD^#	46.82	<0.001	0.815
2-year mortality
Univariate analysis
Pro-adrenomedullin	26.98	<0.001	0.663
BODE	38.03	<0.001	0.711
BOD^#	30.91	<0.001	0.688
Multivariate analysis
Pro-adrenomedullin plus:
BODE	59.46⁺	<0.001	0.771
BOD^#	55.60	<0.001	0.764

Analysed cohort n=594 (549 with all data available plus the 45 (88.2%) out of 51 patients with missing 6-min walk distance but available pro-adrenomedullin data). p-values compare variables with the null hypothesis. BODE: body mass, airflow obstruction, dyspnoea, exercise capacity index. ^#: BOD comprises BODE without the exercise capacity measurement; the three BOD components are scored according to the same cut-offs as they are in BODE [2], therefore, the “BOD index” ranges, in increasing severity, from 0 to 7. ^¶: for 1-year mortality prediction in this sensitivity analysis, value added of pro-adrenomedullin with BODE Chi-squared 19.78, p<0.001. ⁺: for 2-year mortality prediction in this sensitivity analysis, value added of pro-adrenomedullin with BODE Chi-squared 21.43, p<0.001.