Table 2.
Domain | Variable | Feature sets | |||||||||
---|---|---|---|---|---|---|---|---|---|---|---|
Feature set 1: sociodemographic and clinical variables | Feature set 2: PRO variables added to feature set 1 | ||||||||||
Normalized variable importance (%) | Normalized variable importance (%) | ||||||||||
Model | Model | Model | Model | Model LR* | Model | Model | Model | Model | Model LR* | ||
DT | Bagging | RF | AdaBoost | DT | Bagging | RF | AdaBoost | ||||
Clinical factors | Cancer stage II–III | 24.36 | 20.06 | 19.39 | 23.57 | 23.44 | 9.78 | 7.49 | 6.06 | 6.60 | 11.40 |
Local invasion of tumor | 8.90 | 14.34 | 14.28 | 10.66 | 12.50 | 8.20 | 6.31 | 5.58 | 3.26 | NS | |
Regional lymph node metastasis | 23.71 | 10.25 | 10.42 | 9.13 | NS | 8.59 | 6.20 | 6.42 | 7.58 | NS | |
Sociodemographic factors | Low household income (< 3,000$) | 13.82 | 18.46 | 16.00 | 14.26 | 20.49 | 4.93 | 5.33 | 5.29 | 5.60 | 5.14 |
Age over 65 years | 20.45 | 19.87 | 21.87 | 23.19 | 26.75 | 6.04 | 6.28 | 7.40 | 7.02 | 11.61 | |
Male | 8.76 | 17.01 | 18.03 | 19.19 | 24.84 | 5.90 | 5.70 | 7.61 | 6.96 | 11.36 | |
HRQOL factors | BMI (kg/m2) before the operation (≥ 23. 5) | 5.91 | 7.38 | 6.95 | 6.55 | 10.09 | |||||
Anxiety | 3.34 | 5.49 | 7.12 | 6.28 | 7.04 | ||||||
Depression | 3.59 | 6.37 | 6.60 | 6.36 | 4.46 | ||||||
Poor physical functioning | 1.74 | 2.36 | 1.74 | 1.48 | 6.47 | ||||||
Role functioning | 1.64 | 1.73 | 1.53 | 2.14 | 3.54 | ||||||
Poor dyspnea | 5.47 | 5.89 | 6.59 | 7.51 | 3.82 | ||||||
Poor appetite loss | 3.53 | 4.24 | 3.35 | 4.44 | NS | ||||||
Poor diarrhea | 1.95 | 2.64 | 2.10 | 2.78 | NS | ||||||
Poor lung cancer-specific cough | 3.00 | 4.27 | 3.98 | 4.28 | NS | ||||||
Poor pain in chest | 3.36 | 4.41 | 4.34 | 3.88 | NS | ||||||
Low new possibility | 5.93 | 5.26 | 4.88 | 3.69 | 7.69 | ||||||
Low personal strength | 6.62 | 5.45 | 5.56 | 6.11 | 12.23 | ||||||
Low appreciation of life | 10.48 | 7.21 | 6.89 | 7.47 | 5.19 |
NS, nonsignificant; BMI, body mass index; HRQOL, health-related quality of life; DT, decision tree; RF, random forest; LR, logistic regression.
*LR variable selection using stepwise feature selection with a 5% significance level.
The most important variable in the top 20% from each model are highlighted in bold font.