Predictors for mortality due to acute exacerbation of COPD in primary care: Derivation of a clinical prediction rule in a multicentre cohort study

Figures & data

Figure 1. Patient flow diagram. EHR: electronic health record

Table 1. Descriptive characteristics of the sample and their relation to 30-day mortality in the univariate analysis.

Variable (unit)	n (%)^a		Missing data (%)		OR (CI 95%)		p-Value	PC+	PC-
Sex, male	1.419	(83.7)	–		1.46	(0.34 − 6.37)	1	1.06	0.72
Age (years)^b	78	(69 − 83)	–		1.06	(1 − 1.12)	0.048	–	–
Partial saturation of O2 (%)^b	93	(90 − 95)	683	(40.3)	0.88	(0.84 − 0.93)	<0.001	–	–
Respiratory rate (min^-1)^b	20	(15 − 24)	1.596	(94.1)	1.12	(1 − 1.26)	0.043	–	–
Systolic blood pressure (mm Hg)^b	134.5	(120 − 145)	1.302	(76.8)	0.97	(0.95 − 1)	0.072	–	–
Diastolic blood pressure (mm Hg)^b	70	(64 − 80)	1.304	(76.9)	0.96	(0.92 − 1.01)	0.136	–	–
Heart rate (min^-1)^b	84	(72 − 95)	1.023	(60.3)	1.03	(1.01 − 1.06)	0.014	–	–
Temperature (°C)^b	36.4	(35.9 − 37.2)	1.433	(84.5)	1.04	(0.5 − 2.16)	0.924	–	–
Oedema	93	(5.5)	1.410	(83.1)	1.77	(0.53 − 5.96)	0.345	1.42	0.8
Retractions or use of accessory respiratory muscles	46	(2.7)	1.551	(91.5)	3.38	(0.55 − 20.99)	0.327	1.95	0.58
Confusion	8	(0.5)	479	(28.2)	30.7	(5.65 − 166.37)	0.004	26.31	0.87
Degree of dyspnoea (mMRC)	–		1.444	(85.1)	–		–	–	–
Dyspnoea grade 0	48	(2.8)	–		–		–	–	–
Dyspnoea grade 1	11	(0.6)	–		–		–	–	–
Dyspnoea grade 2	38	(2.2)	–		–		–	–	–
Dyspnoea grade 3	42	(2.5)	–		–		–	–	–
Dyspnoea grade 4	113	(6.7)	–		9.11	(1.1 − 75.21)^c	0.024	2.04	0.22
%FEV1*	56.4	(48.3 − 68.7)	1.672	(98.6)	–		–	–	–
Body mass index (kg/m^2)^b	28.3	(25.6 − 31.6)	739	(43.6)	0.9	(0.75 − 1.09)	0.283	–	–
Exacerbations in the last 12 months^b	1	(0 − 2)	–		1.57	(1.2 − 2.05)	0.001
Charlson Index^b	1	(1 − 2)	–		1.08	(0.88 − 1.33)	0.478	–	–
Charlson Index >1	842	(49.6)	–		2.45	(0.86 − 7)	0.083	1.43	0.58
Cardiovascular disease	363	(21.4)	–		3.31	(1.27 − 8.65)	0.016	2.23	0.67
Peripheral arterial disease	123	(7.3)	–		0.8	(0.1 − 6.06)	1	0.81	1.01
Heart failure	115	(6.8)	–		4.35	(1.39 − 13.55)	0.024	3.56	0.82
Cerebrovascular disease	102	(6)	–		2.1	(0.47 − 9.33)	0.273	1.98	0.94
Acute myocardial infarction	79	(4.7)	–		4.52	(1.27 − 16.06)	0.041	3.9	0.86
Diabetes mellitus	400	(23.6)	–		1	(0.32 − 3.07)	1	1	1
Dementia	24	(1.4)	–		–		1	–	–
Cancer^d	162	(9.6)	–		2.05	(0.58 − 7.2)	0.217	1.86	0.91
Home care programme	110	(6.5)	–		1.94	(0.44 − 8.59)	0.303	1.83	0.94
Rural health centre	773	(46.6)	–		2.21	(0.81 − 5.99)	0.142	1.43	0.65
Season	–		–		–		–	–	–
Spring	390	(23.0)	–		1.84	(0.68 − 5.01)^e	0.246	1.54	0.84
Summer	236	(13.9)	–		1.33	(0.38 − 4.66)^e	0.721	1.27	0.96
Autumn	520	(30.7)	–		0.94	(0.33 − 2.69)^e	1	0.96	1.02
Winter	550	(32.4)	–		0.44	(0.13 − 1.55)^e	0.297	0.54	1.22
Death at 30 days	17	(1)	–		–		–	–	–

CI: confidence interval; PC+: positive likelihood ratio; PC-: negative likelihood ratio; O_2: oxygen; min: minute; mmHg: millimetres of mercury; °C: degrees Celsius, %FEV₁: percentage of forced expiratory volume in the first second in relation to the expected FEV; mMRC: modified dyspnoea scale of the Medical Research Council.

a if not otherwise specified.

b Median (interquartile range).

c dyspnoea = 4 vs. dyspnoea <4.

d except basal cell carcinoma.

e vs. rest of stations.

Figure 2. Calibration plot. Solid line represents a model with perfect calibration. Dashed line represents a non-parametric smooth curve for the relation between observed frequency and predicted probability. Triangles are based on quintiles of patients with similar predicted probabilities. Vertical lines above the x axis represents the distribution of predicted probabilities.

Figure 3. Receiver operating characteristic curve. Dashed line is the no discrimination line (AUROC: 0.5). Solid line represents the model developed (AUROC: 0.811 (95% CI, 0.72–0.902)).

Table 2. Characteristics of each risk category and associated recommendation.

Risk category	Predicted probability	LR+	LR-	Net benefit	Proportion of individuals (%)	Proposed action
Low	<0.008	2.51	0.19	0.0035	40	Probability of death is below prevalence. Consider treating the patient in primary care
Medium	0.008 − 0.029	–	–		20	Probability of death is higher than prevalence. If you decide to treat the patient in primary care, follow him/her closely
High	≥0.03	5.32	0.91	0.0077	20	Probability of death is in the top quintile. Consider referring the patient to the hospital

LR+: positive likelihood ratio; LR-: negative likelihood ratio.

Figure 4. Decision curve. Solid black line represents ‘no intervention’ (derive all). Solid grey line represents the net benefit for ‘intervention in all’ (derive none). Dashed line represents the net benefit for the model.

Table 3. EXAGGERATE score.

Predictor	Points
Exacerbations in the last 12 months	1 point/exacerbation
Age >75 years	1 point
Heart rate >100 min^-1	1 point
Risk category	Score
Low	0 − 1 points
Medium	2 − 3 points
High	≥4 points