Table 2

Performance metrics for the internal cross-validation and external validation of the COPD risk-prediction model, with and without imputation in the development TAHS dataset*

Model validation (n/N=COPD/total cases)Diagnostic metrics (SE) †HL 2 p value
AUCROC(Cut-off)‡SensSpecNPVPPV
Internal validation (TAHS)
Complete case model (n/N=106/2320)0.808 (0.004)0.4800.803 (0.013)0.691 (0.002)0.987 (0.001)0.111 (0.004)0.13
0.500.779 (0.017)0.713 (0.004)0.985 (0.001)0.115 (0.002)
Imputed data model (n/N=108/2407)0.811 (0.004)0.4500.816 (0.013)0.671 (0.003)0.987 (0.001)0.105 (0.002)0.30
0.500.764 (0.012)0.724 (0.003)0.985 (0.001)0.115 (0.002)
External validations (ECRHS) using complete case model§
Equivalent age group (n/N=39/548)§0.746 (0.006)0.4830.745 (0.010)0.668 (0.003)0.972 (0.001)0.148 (0.005)0.95
0.500.666 (0.011)0.686 (0.003)0.964 (0.001)0.141 (0.005)
Extended age group (n/N=95/1407)¶0.756 (0.001)0.4830.769 (0.003)0.659 (0.001)0.975 (0.001)0.140 (0.001)0.69
0.500.737 (0.003)0.677 (0.001)0.975 (0.001)0.142 (0.001)
Current smokers only (n/N=36/268)0.639** (0.010)0.500.835 (0.011)0.173 (0.003)0.870 (0.008)0.137 (0.003)
Current asthma only (n/N=32/142)0.458** (0.006)0.500.969 (0.005)0.018 (0.002)0.662 (0.055)0.223 (0.005)
Any current respiratory symptom (n/N=72/631)0.719** (0.004)0.500.905 (0.005)0.469 (0.003)0.975 (0.001)0.179 (0.004)
  • *In TAHS, complete case numbers (n/N=106/2320) and imputed data (n/N=108/2407 participants).

  • †SE=SD deviations from the mean (equivalent to SE).

  • ‡Based on >50% predicted probably for a positive case or optimised cut-off as per the Youden index.

  • §Data from ECRHS II (age 40–44) and ECRHS III (age 50–55) (n/n=39/548 participants).

  • ¶Data from ECRHS II (age 40–49) and ECRHS III (age 50–59) (n/n=95/1407 participants).

  • **AUCROC values based only on a subset of data are poor indicators of model performance (as not based on the entire dataset).

  • AUCROC, area under the receiver operator characteristic curve; COPD, chronic obstructive lung disease; ECRHS, European Community Respiratory Health Survey; HL, Hosmer-Lemeshow; n, number of COPD cases; N, total number; NPV, negative predictive values; PPV, positive predictive value; sens, sensitivity; spec, specificity; TAHS, Tasmanian Longitudinal Health Study.