Table 3– Summary of models for detecting severe chronic obstructive pulmonary disease (COPD): validation sample

Prediction variables used in the model	Definition of group	Population %	Cases^# %	Risk
General (unstaged) CART modelling
Questionnaire data only
High risk	Patient reports doctor-diagnosed COPD/emphysema or dyspnoea ≥grade 3 or BMI <20 kg·m⁻² or wheeze with shortness of breath	19.6	70.0	11.4
Low risk	Everyone else	80.4	30.0	1.2
Pre-BD peak flow and questionnaire data
High	PEF per Ht² ≤1.8 L·s⁻¹·m⁻²	9.0	93.3	33.3
Low	PEF per Ht² >1.8 L·s⁻¹·m⁻²	91.0	6.7	0.2
Pre- and post-BD peak flow and questionnaire data
High	Pre-BD PEF per Ht² ≤1.3 L·s⁻¹·m⁻² or post-BD PEF per Ht² ≤1.8 L·s⁻¹·m⁻²	7.5	92.2	39.3
Low	Pre-BD PEF per Ht² >1.3 L·s⁻¹·m⁻² and post-BD PEF per Ht² >1.8 L·s⁻¹·m⁻²	92.5	7.8	0.3
Staged screening models
Questionnaire data only
High risk	Symptomatic	73.8	97.8	4.2
Low risk	Asymptomatic	26.2	2.2	0.3
Pre-BD peak flow and questionnaire data
High risk	Symptomatic and PEF per Ht² ≤1.8 L·s⁻¹·m⁻²	8.1	91.1	36.0
Low risk	Asymptomatic or PEF per Ht² >1.8 L·s⁻¹·m⁻²	91.9	8.9	0.3
Pre- and post-BD peak flow and questionnaire data
High risk	Symptomatic and pre-BD PEF per Ht² ≤1.8 L·s⁻¹·m⁻²	8.1	91.1	36.0
Low risk	Asymptomatic or pre-BD PEF per Ht² >1.8 L·s⁻¹·m⁻²	91.9	8.9	0.3

Data for risk are presented as positive predictive value. CART: classification and regression tree; BD: bronchodilator; BMI: body mass index; PEF: peak expiratory flow; Ht: height. ^#: sensitivity.