Comparison of model-building strategies for excess hazard regression models in the context of cancer epidemiology

Maringe, Camille; Belot, Aurélien; Rubio, Francisco Javier; Rachet, Bernard

doi:10.1186/s12874-019-0830-9

BMC Medical Research Methodology

Table 2 Summary of models and variables selected by each algorithm, on 250 samples of N = 2000 and N = 5000 patients: scenarios A-D

From: Comparison of model-building strategies for excess hazard regression models in the context of cancer epidemiology

N = 2000
	Overall model									Sensitivity			Specificity
	Contained			Correctly selected			Almost correctly selected*			mean	min	max	mean	min	max
A	p (%)	95% CI**		p (%)	95% CI**		p (%)	95% CI**
MVRS	100.0	100	100	69.6	64.0	75.2				0.97	0.67	1.00	0.89	0.75	0.92
W&A	100.0	100	100	70.4	64.8	76.0				0.98	0.67	1.00	0.89	0.67	0.92
C
MVRS	10.8	7.0	14.6	8.8	5.3	12.3	82.8	78.2	87.4	0.74	0.60	0.80	0.88	0.70	0.90
W&A	6.0	3.1	8.9	5.2	2.5	7.9	91.6	88.2	95.0	0.74	0.60	0.80	0.88	0.60	0.90
B
aMVRS	35.6	29.8	41.5	14.4	10.1	18.7	53.6	47.5	59.7	0.80	0.50	1.00	0.85	0.45	0.91
MFPIgen	29.6	24.0	35.2	14.4	10.1	18.7	66.8	61.1	72.6	0.81	0.50	1.00	0.88	0.64	0.91
aW&A	35.2	29.4	41.0	14.8	10.5	19.1	45.2	39.1	51.3	0.79	0.50	1.00	0.84	0.36	0.91
D
aMVRS	2.4	0.5	4.3	1.6	0.1	3.1	16.0	11.5	20.5	0.56	0.50	0.83	0.80	0.33	0.89
MFPIgen	3.2	1.1	5.4	2.8	0.8	4.8	36.8	30.9	42.7	0.59	0.50	0.83	0.85	0.56	0.89
aW&A	4.8	2.2	7.4	1.6	0.1	3.1	23.2	18.1	28.4	0.57	0.50	0.83	0.75	0.22	0.89
N = 5000
	Overall model									Sensitivity			Specificity
	Contained			Correctly selected			Almost correctly selected*			mean	min	max	mean	min	max
A	p (%)	95% CI**		p (%)	95% CI**		p (%)	95% CI**
MVRS	100.0	100	100	56.4	50.4	62.5				0.97	0.67	1.00	0.88	0.67	0.92
W&A	100.0	100	100	68.0	62.3	73.7				0.97	0.67	1.00	0.89	0.67	0.92
C
MVRS	46.0	39.9	52.1	40.8	34.8	46.8	78.8	73.8	83.8	0.78	0.60	0.80	0.88	0.60	0.90
W&A	29.2	23.7	34.8	26.0	20.7	31.4	87.6	83.6	91.6	0.80	0.60	0.80	0.88	0.60	0.90
B
aMVRS	67.9	62.2	73.6	37.3	31.5	43.3	55.8	49.7	61.9	0.80	0.50	1.00	0.87	0.64	0.91
MFPIgen	28.0	22.5	33.5	14.4	10.1	18.7	65.6	59.8	71.4	0.87	0.50	1.00	0.85	0.36	0.91
aW&A	69.2	63.6	74.8	37.6	31.7	43.5	55.6	49.5	61.7	0.86	0.50	1.00	0.82	0.09	0.91
D
aMVRS	34.4	28.6	40.2	13.6	9.4	17.8	24.8	19.5	30.1	0.66	0.33	0.83	0.79	0.33	0.89
MFPIgen	28.0	22.5	33.5	18.4	13.7	23.1	35.6	29.8	41.5	0.69	0.50	0.83	0.84	0.56	0.89
aW&A	22.0	17.0	27.1	13.6	9.4	17.8	34.8	29.0	40.6	0.65	0.50	0.83	0.73	0.00	0.89

* model C: relaxed NL and TD of age; B: relaxed interaction age*stage; D: relaxed NL and TD of age
** formula for the 95% confidence intervals, with z = 1.96 and w = 250: \( \frac{\hat{p}+\frac{z^2}{2w}}{1+\frac{z^2}{w}}\pm \frac{z}{1+\frac{z^2}{w}}\sqrt{\frac{\hat{p}\left(1-\hat{p}\right)}{w}+\frac{z^2}{4{w}^2}} \), using the Wilson approximation [36]

Back to article page

ISSN: 1471-2288

Contact us

General enquiries: journalsubmissions@springernature.com