Table 5

Comparison of mortality and person-years lost for data-derived and constrained models in development and validation cohorts

All cause mortality

Breast specific mortality

Person-Years Lost

ECRIC

Model:

Hormone coefficient

Chemotherapy coefficient

A

P

ROC

A

P

ROC

Total Possible PY

A

P


Full

Model

Model

1,075

1,082

0.81

737

768

0.84

35,003

3,099

3,341

Full

Overview*

Overview

1,075

984

0.81

737

660

0.83

35,003

3,099

3,030

Constrained

Model

Model

1,075

980

0.81

737

656

0.83

35,003

3,099

3,061

Constrained

Overview

Overview

1,075

990

0.81

737

667

0.83

35,003

3,099

3,089

WMCIU

Model:

Full

Model

Model

955

1,006

0.79

668

743

0.82

28,322

2,405

2,513

Full

Overview

Overview

955

956

0.79

668

690

0.81

28,322

2,405

2,376

Constrained

Model

Model

955

952

0.78

668

685

0.81

28,322

2,405

2,406

Constrained

Overview

Overview

955

965

0.78

668

699

0.81

28,322

2,405

2,436


*Application of overview estimates is performed by setting the model-derived therapy coefficient to zero. Overview-derived therapy risk reductions (first generation) are then applied in the life table analysis. The risk reductions are only applied to individuals who actually received therapy. No Overview hormone therapy benefit was applied to individuals with ER- tumors.

Under the constrained model, the model coefficient for adjuvant hormone therapy is the same as the Overview estimate.

Follow-up was censored at eight years

Wishart et al. Breast Cancer Research 2010 12:R1   doi:10.1186/bcr2464

Open Data