- Email: [email protected]
Validity of Adjuvant! Online program in older patients with breast cancer: a population-based study
Articles
Validity of Adjuvant! Online program in older patients with breast cancer: a population-based study Nienke A de Glas, Willemien van de Water, Ellen G Engelhardt, Esther Bastiaannet, Anton J M de Craen, Judith R Kroep, Hein Putter, Anne M Stiggelbout, Nir I Weijl, Cornelis J H van de Velde, Johanneke E A Portielje, Gerrit-Jan Liefers
Summary Background Adjuvant! Online is a prediction tool that can be used to aid clinical decision making in patients with breast cancer. It was developed in a patient population aged 69 years or younger, and subsequent validation studies included small numbers of older patients. Since older patients with breast cancer differ from younger patients in many aspects, the aim of this study was to investigate the validity of Adjuvant! Online in a large cohort of unselected older patients. Methods We included patients from the population-based FOCUS cohort, which included all consecutive patients aged 65 years or older who were diagnosed with invasive or in-situ breast cancer between Jan 1, 1997, and Dec 31, 2004, in the southwestern part of the Netherlands. We included all patients who fulfilled the criteria as stated by Adjuvant! Online: patients with unilateral, unicentric, invasive adenocarcinoma; no evidence of metastatic or residual disease; no evidence of T4 features; and no evidence of inflammatory breast cancer. We entered data from all patients with the “average for age” comorbidity status (model 1) and with an individualised comorbidity status (model 2). Findings We included 2012 patients. Median age of patients in the cohort was 74·0 years (IQR 69·0–79·0). 904 (45%) of 2012 patients died during follow-up, whereas 326 (16%) patients had recurrence. Median follow-up for overall survival was 9·0 years (IQR 7·4–10·7), and 6·6 years (4·4–6·6) for patients without recurrence. Using model 1, Adjuvant! Online overestimated 10-year overall survival by 9·8% ([95% CI 5·9–13·7], p<0·0001) and 10-year cumulative recurrence survival by 8·7% ([6·7–10·7], p<0·0001). By contrast, when using model 2, Adjuvant! Online underestimated the 10-year overall survival by –17·1% ([95% CI –21·0 to –13·2], p<0·0001). However, when using model 2, Adjuvant! Online predicted cumulative recurrence accurately in all patients (–0·7% [95% CI –2·7–1·3], p=0·48). Interpretation Adjuvant! Online does not accurately predict overall survival and recurrence in older patients with early breast cancer. Funding Dutch Cancer Foundation.
Introduction Adjuvant systemic treatment in early stage breast cancer is aimed at preventing breast cancer recurrence and mortality. The prognosis for early stage breast cancer has greatly improved with the introduction of adjuvant treatments.1 Optimum combinations of adjuvant endocrine therapy and chemotherapy now result in relative risk reductions of between 20% to 57% in 15-year mortality.1 In patients with a high absolute risk of recurrence, the potential benefit of adjuvant treatment is large. However, the benefit of adjuvant therapy might be attenuated in the presence of increased comorbidities or old age because of shorter life expectancy and competing causes of death.1–3 Therefore, the correct identification of patients who have an increased absolute risk of breast cancer recurrence and breast cancer mortality, and who might benefit from adjuvant therapy, is essential. In clinical practice, many prediction methods are used to estimate the benefit of adjuvant treatment. Of these prediction methods Adjuvant! Online is best known.4 Adjuvant! Online is an online, open-access prediction program that predicts 10-year breast cancer recurrence, breast cancer mortality, mortality due to other causes,
and expected benefits of specific adjuvant treatment options for individual patients.4 Predictions are based on six clinical factors, including comorbidity. The model was developed using a large database that was derived from the Surveillance, Epidemiology, End-results (SEER) registry, with a population of 34 252 patients.4 The study population consisted of women aged 35–69 years, who were diagnosed with early breast cancer between 1988 and 1992 in the USA.5 Adjuvant! Online has been validated in Canada, the USA, and in several Asian and European studies.5–11 Although Adjuvant! Online was developed in a cohort with quite young patients, the Dutch national guideline recommends using Adjuvant! Online for decision making in older patients (those aged 70 years or older) with breast cancer, on the basis of the original Canadian validation study.5,12 Treatment recommendations of the International Society of Geriatric Oncology from 2007 included using Adjuvant! Online in clinical decision making, although the most recent update did not mention the use of Adjuvant! Online.13,14 However, older patients make up a heterogeneous group, partly because of the presence of additional comorbidities.15,16 Survival is
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Lancet Oncol 2014 Published Online May 14, 2014 http://dx.doi.org/10.1016/ S1470-2045(14)70200-1 Department of Surgery (N A de Glas MD, W van de Water MD, E Bastiaannet PhD, Prof C J H van de Velde PhD, G-J Liefers PhD), Department of Gerontology and Geriatrics (N A de Glas, W van de Water, E Bastiaannet, A J M de Craen PhD), Department of Medical Decision Making (E G Engelhardt MSc, Prof A M Stiggelbout PhD), Department of Medical Oncology (J R Kroep PhD), Department of Medical Statistics (Prof Hein Putter PhD), Leiden University Medical Centre, Leiden, Netherlands; Department of Medical Oncology, Bronovo Hospital The Hague, The Hague, Netherlands (N I Weijl PhD); and Department of Medical Oncology, Haga Hospital The Hague, Leyweg The Hague, Netherlands (J E A Portielje PhD) Correspondence to: Dr Gerrit-Jan Liefers, Leiden University Medical Centre, Department of Surgery, Postzone K6-R, PO Box 9600, 2300 RC Leiden, Netherlands [email protected]
1
Articles
shorter in older patients with breast cancer compared with younger patients with breast cancer, mainly because the risk of competing mortality increases with age.3,17 Furthermore, older patients were under-represented in the trials which were used to predict the benefits of adjuvant therapy.1 This might hamper the generalisability of trial results to older patients with breast cancer. These factors probably affect predictions made by Adjuvant! Online. Additionally, results from a recent systematic review showed that the validity of its estimates has not been assessed with sufficient power in older patients.11 The aim of this study was to investigate the validity of Adjuvant! Online by assessing its discriminatory accuracy and calibration in a large cohort of women aged 65 years or older diagnosed with in-situ and invasive breast cancer.
Methods Patients and study design We used data from the population-based FOCUS cohort.18,19 This cohort includes all patients with incident breast cancer aged 65 years or older who were diagnosed in the geographically defined Comprehensive Cancer Center Region West in The Netherlands between Jan 1, 1997, and Dec 31, 2004 (n=3672). There were no exclusion criteria. Patients were identified through the Netherlands cancer registry, which registers information about all patients with cancer in the Netherlands through the central pathology database. Trained personnel reviewed the charts of all identified patients, and obtained information on specific treatments, comorbidity according to the ICD-10 classification,20 adverse events, geriatric variables (difficulty walking, sensory handicap, dementia, polypharmacy, and nursing home resident), and recurrences. Follow-up mortality data was available until Jan 1, 2011, by linking cancer registry data with municipal population registries. In the present study, all patients who fulfilled the criteria as stated by Adjuvant! Online were included: patients with unilateral, unicentric, invasive adenocarcinoma; no evidence of metastatic or residual disease; no evidence of T4 features; and no evidence of inflammatory breast cancer. Furthermore, patients had to have had previous definitive primary breast surgery and axillary node staging; not received neoadjuvant systemic treatment or radiotherapy; and be treated with radiotherapy if the patient had a lumpectomy.
Procedures We entered patient and tumour characteristics in the Adjuvant! Online program version 8·0, to calculate predicted 10-year overall survival and 10-year cumulative recurrence for every patient. We entered age as a continuous variable. In Adjuvant! Online, comorbidity is classified as “perfect health”; “minor problems”; “average for age”; “major problems+10”; “major problems+20”; and “major problems+30”, although cutoff points or 2
definitions of these different groups are not provided. Therefore, we used two different classifications of comorbidity. First, we entered the “average for age” category for all patients, as we expected that for all patients together, this category will give average outcomes of the predicted recurrence and survival rates that are most likely to portray the general population (model 1). Second, we assembled a panel consisting of one medical oncologist, one dedicated geriatric oncologist, one surgical oncologist, and three epidemiologists. The panel specified the categories of comorbidity on the basis of the effect of specific conditions on mortality, daily functioning, and quality of life (model 2). Patients were classified on the basis of the worst diagnosed condition. All analyses were done for both models. In Adjuvant! Online, pathological tumour size is classified as 0–1·0 cm, 1·1–2·0 cm, 2·1–3·0 cm, 3·1–5·0 cm, or larger than 5·0 cm. Exact tumour size was missing in 86 (4%) patients from our data (n=2012); for these patients, we used pathological T-stage instead. Since the categories of tumour size of Adjuvant! Online do not exactly resemble TNM-stage, we classified these 86 patients in the category that mostly resembled the T-stage as follows. Tumour size of patients with stage T1 was recoded as 1·1–2·0 cm, stage T2 was recoded as 3·1–5·0 cm, and stage T3 was recoded as larger than 5·0 cm. In Adjuvant! Online, tumour grade is defined as grade I, II, or III, or “undefined” in case of missing values. Nodal stage is classified as zero positive nodes, one to three positive nodes, four to nine positive nodes, or more than nine positive nodes. The exact number of positive nodes was missing in 298 (15%) of 2012 patients, as this information was often not registered in the pathology report. These patients with missing data for the number of positive axillary lymph nodes were classified using pathological N-stage (if available) or clinical N-stage as follows: N0 was recoded as no positive nodes, stage N1 was recoded as one to three positive nodes, stage N2 was recoded as four to nine positive nodes, and stage N3 was recoded as more than nine positive nodes. Finally, oestrogen receptor (ER) status was defined as positive, negative, or “undefined” in case of missing values. We used the definition of ER positivity that was given by the pathologist who rated the tumour sample. We assumed that missing data about both tumour grade and ER-status were completely missing at random. Whether patients received adjuvant endocrine therapy or chemotherapy was registered, but individual information on the specific treatment regimen was not available. Therefore, we defined the type of adjuvant therapy in line with the most commonly used treatments at the time of diagnosis. We defined endocrine therapy, if given, as tamoxifen only for patients who were diagnosed between 1997 and 1999 and as sequential therapy for patients who were diagnosed between 2000 and 2004. We
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
defined chemotherapy, if given, as first generation therapy for all patients.
Statistical analysis Primary outcome measures were 10-year overall survival and 10-year cumulative recurrence, which was defined as
n (%)
locoregional recurrence, distant recurrence, or contralateral breast cancer, in line with the definition of Adjuvant! Online. We first assessed the association between our comorbidity classification and overall survival to establish the validity of the classification using a Cox regression model. The model was adjusted for age only, since age was
10-year overall survival
10-year cumulative recurrence
Adjuvant predicted
Observed (SE)
Predicted–observed (95% CI)
2012 (100%)
48·8
39·0 (2)
65–69
560 (28%)
66·9
70–74
544 (27%)
57·9
75–79
417 (21%)
80–84
322 (16%)
≥85
169 (8%)
All patients
p value
Adjuvant predicted
Observed (SE)
Predicted–observed (95% CI)
p value
9·8 (5·9 to 13·7)
<0·0001
26·9
18·2 (1·0)
8·7 (6·7 to 10·7)
61·0 (3)
5·9 (0·01 to 12·8)
0·05
27·1
20·1 (2·0)
7·0 (3·1 to 10·9)
0·0005
51·0 (4)
6·9 (–1·0 to 14·8)
0·08
28·0
17·1 (1·8)
10·9 (7·4 to 14·4)
<0·0001
44·4
26·0 (3)
18·4 (12·5 to 24·3)
<0·0001
27·6
20·4 (2·2)
7·2 (2·9 to 11·5)
0·001
27·1
16·0 (4)
11·1 (3·2 to 19·0)
0·006
27·1
17·8 (2·4)
9·3 (4·6 to 14·0)
0·0001
11·3
5·0 (3)
6·3 (0·4 to 12·2)
0·04
20·8
9·1 (2·3)
11·7 (7·2 to 16·2)
<0·0001
<0·0001
Age (years)
Number of comorbidities 0
485 (24%)
52·2
55·0 (3)
–2·8 (–8·7 to 3·1)
0·35
27·4
19·7 (2·1)
7·7 (3·6 to 11·8)
0·0002
1
476 (24%)
50·9
50·0 (4)
0·9 (–7·0 to 8·8)
0·82
28·0
17·3 (2·0)
10·1 (6·8 to 14·6)
<0·0001
2
398 (20%)
47·3
33·0 (4)
14·3 (6·4 to 22·2)
<0·0001
26·2
17·1 (2·0)
9·1 (5·2 to 13·0)
0·0008
3
275 (14%)
47·4
31·0 (5)
16·4 (6·6 to 27·2)
0·001
25·9
17·1 (2·7)
8·8 (3·5 to 14·1)
0·001
≥4
378 (19%)
44·5
19·0 (3)
25·5 (19·6 to 31·4)
<0·0001
26·6
19·3 (2·2)
7·3 (3·0 to 11·6)
0·0009
Oestrogen receptor status Positive
1404 (79%)
42·8
37·0 (2)
5·8 (1·9 to 9·7)
0·004
24·4
17·7 (1·2)
6·7 (4·3 to 9·1)
<0·0001
Negative
370 (29%)
49·9
39·0 (4)
10·9 (3·0 to 18·8)
0·006
36·1
22·9 (2·2)
13·2 (8·9 to 17·5)
<0·0001
Undefined*
238
51·0
43·0 (4)
8·0 (0·1 to 15·9)
0·05
27·3
13·9 (2·5)
13·4 (8·5 to 18·3)
<0·0001
Tumour grade Grade 1
293 (20%)
57·2
45·0 (5)
12·2 (2·4 to 22·0)
0·01
17·2
10·2 (3·5)
7·0 (0·1 to 13·9)
0·05
Grade 2
682 (58%)
50·2
39·0 (4)
11·2 (3·4 to 19·1)
0·005
24·1
15·2 (1·7)
8·9 (5·6 to 12·2)
<0·0001
Grade 3
459 (22%)
43·6
35·0 (4)
8·6 (0·7 to 16·5)
0·03
33·9
29·0 (2·5)
Undefined*
578
46·8
40·0 (3)
6·8 (0·9 to 12·7)
0·02
29·7
18·1 (1·7)
0·1–1·0
299 (15%)
65·1
56·0 (5)
9·1 (–0·7 to 18·9)
0·07
17·0
8·5 (2·3)
8·5 (4·0 to 13·0)
0·0002
1·1–2·0
810 (40%)
54·8
45·0 (3)
9·8 (3·9 to 15·7)
0·001
21·2
14·9 (1·4)
6·3 (3·6 to 9·0)
<0·0001
2·1–3·0
540 (27%)
40·4
29·0 (3)
11·4 (5·5 to 17·3)
0·0001
32·3
23·6 (2·0)
8·7 (4·8 to 12·6)
<0·0001
3·1–5·0
301 (15%)
35·1
32·0 (4)
3·1 (–4·8 to 11·0)
0·44
37·4
24·9 (2·8)
12·5 (7·0 to 18·0)
<0·0001
62 (3%)
29·7
21·0 (7)
8·7 (–5·3 to 22·7)
0·21
51·4
28·6 (5·9)
22·8 (11·0 to 34·6)
0·0001
4·9 (–0·01 to 9·8) 11·6 (8·3 to 14·9)
0·05 <0·0001
Tumour size (cm)
>5 Positive nodes 0
1385 (69%)
53·9
43·0 (2)
10·9 (7·0 to 14·8)
<0·0001
21·8
13·6 (1·1)
8·2 (6·0 to 10·4)
<0·0001
1–3
475 (24%)
40·4
34·0 (3)
6·4 (0·5 to 12·3)
0·03
33·3
24·4 (2·3)
8·9 (4·4 to 13·4)
0·0001
4–9
100 (5%)
31·2
18·0 (5)
13·2 (3·3 to 23·1)
0·008
47·0
36·7 (5·0)
10·3 (0·4 to 20·2)
0·04
>9
52 (3%)
21·2
24·0 (9)
–2·8 (–20·9 to 15·3)
0·76
66·5
44·0 (7·5)
22·5 (7·4 to 37·6)
0·003
781 (39%)
60·4
55·0 (3)
5·4 (–0·5 to 11·3)
0·07
23·8
14·6 (1·4)
9·2 (6·5 to 12·0)
<0·0001
1231 (61%)
41·4
29·0 (2)
12·4 (8·5 to 16·3)
<0·0001
28·9
20·5 (1·3)
8·4 (5·9 to 11·0)
<0·0001
1133 (56%)
51·6
40·0 (2)
11·6 (7·7 to 15·5)
<0·0001
24·9
13·4 (1·1)
11·5 (9·3 to 13·7)
<0·0001
778 (39%)
44·0
38·0 (3)
6·0 (0·1 to 11·9)
0·05
28·9
23·4 (1·8)
5·5 (2·0 to 9·0)
0·002
Chemotherapy only
52 (3%)
52·2
35·0 (13)
17·2 (–8·9 to 43·3)
0·19
36·9
18·3 (5·6)
18·6 (7·4 to 29·8)
0·001
Hormonal therapy and chemotherapy
49 (2%)
53·2
43·0 (11)
10·2 (–11·9 to 32·3)
0·36
32·5
56·7 (11·5)
24·2 (–1·1 to 47·3)
0·04
Most extensive surgery type Breast conserving surgery Mastectomy Systemic treatment None Hormonal therapy only
ER=oestrogen receptor. *The category “undefined” was not included in the given percentages.
Table: Baseline characteristics and predicted versus observed 10 year overall survival and cumulative recurrence (model 1)
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
3
Articles
likely to be the only confounder in the relation between the comorbidity classification and overall survival. Second, we calculated observed 10-year overall survival using life tables derived from the Kaplan-Meier method.21 We calculated observed 10-year cumulative recurrence using cumulative incidence rates of breast cancer recurrence. Cause-specific outcomes might be affected by the risk of competing endpoints, particularly in older populations.22 Therefore, we estimated cumulative breast cancer recurrence by regression analyses according to Fine and Gray to assess the risk of a breast cancer recurrence while taking into account the risks of reaching other, competing endpoints (death without recurrence).23
See Online for appendix
A Model 1 Model 2 Perfect line (x=y)
100
p<0·0001
Observed 10-year overall survival (%)
80
60 p<0·0001 40
20
We compared the observed and predicted 10-year overall survival and cumulative recurrence using onesample Z-tests, with the predicted survival as fixed value and the observed value as the assessed variable. Subgroups were based on the information requested by Adjuvant! Online to predict outcomes as described here. Additionally, we calculated observed and predicted outcomes for subgroups based on the number of comorbidities (0, 1, 2, 3, and 4 or more), most extensive surgery (breast conserving surgery and mastectomy), and systemic treatment (none, endocrine therapy only, chemotherapy only, endocrine therapy, and chemotherapy). Furthermore, we assessed the calibration of the model by plotting the difference between observed and predicted outcomes for overall survival and cumulative recurrence. For this, we grouped patients into 10% intervals of the predicted 10-year overall survival. Intervals with fewer than 100 patients were combined. We tested regression lines and corresponding p values of the calibration plot against the so-called ideal line (x=y) using Poisson regression models.24 Additionally, we assessed the discriminatory accuracy by composing receiver-operator curves (ROC) and corresponding c-indices. C-indices are derived by calculating the area under the curve (AUC) of the ROC. A c-index of 1·0 means that the model is perfectly accurate, whereas a c-index of 0·50 means that the model does not predict better than chance.25 To investigate the added value of the prediction made by Adjuvant! Online, we additionally computed c-indices that were derived by using age as sole predictor. We did all analyses in SPSS version 20.0 and R version 2.15.3.
0 0
20
40
60
80
100
Adjuvant! Online predicted 10-year overall survival (%)
The FOCUS project and this analysis were funded by the Dutch Cancer Foundation (KWF 2007-3968). The Dutch Cancer Foundation was not involved in study design, collection, analysis, interpretation of the data, writing the report, nor the decision to submit the paper for publication. The corresponding author had full access to all data and had final responsibility for the decision to submit for publication.
B
Observed 10-year cumulative recurrence (%)
100
80
60
Results p<0·0001
40
p<0·0001 20
0 0
20
40
60
80
Adjuvant! online predicted 10-year cumulative recurrence (%)
Figure 1: Observed versus predicted 10-year overall survival (A) and cumulative recurrence (B)
4
Role of the funding source
100
We included 2012 patients who fulfilled the criteria as stated by Adjuvant! Online (appendix). Median age was 74·0 years (IQR 69·0–79·0), and 1527 (76%) of 2012 patients had one or more comorbidities (table). 904 (45%) of 2012 patients died during follow-up, and 326 (16%) patients developed a recurrence. Median follow-up was 9·0 years (IQR 7·4–10·7) for overall survival, and 6·6 years (IQR 4·4–6·6) for patients without disease recurrence. Of all 896 patients with complete 10-year follow-up, 177 (20%) patients had a recurrence, of which 131 (74%) occurred within the first 5 years.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
These findings contrast with previous validation studies that were done in younger patients with breast cancer. Most studies included very few patients older than 65 years of age,7–10 but some did report age-specific differences.5,6 The initial validation study concluded that breast cancer-specific survival was not accurately predicted in patients aged 76 years and older,5 which is confirmed in the current study. A European validation study including 817 patients aged 70 years and older concluded that although overall survival was not accurately predicted by Adjuvant! Online in patients aged 70 years and older, breast cancer-specific survival was.6 However, sample sizes were small compared with those in the present study and the authors used data from a selected group of patients from a tertiary centre. A
1·0
Model 1 Model 2 Age Reference line
0·8
Sensitivity
0·6
0·4
0·2
C-statistics Model 1: 0·75 (95% CI 0·72–0·77) Model 2: 0·70 (95% CI 0·68–0·73) Age: 0·70 (95% CI 0·68–0·73)
0
B
1·0
0·8
0·6 Sensitivity
The table shows both predicted and observed 10-year overall survival and 10-year cumulative recurrence according to model 1. Using model 1, Adjuvant! Online overestimated the 10-year overall survival by 9·8% ([95% CI 5·9–13·7], p<0·0001). The difference between observed and predicted 10-year overall survival was smallest in patients aged 65–69 years, and largest in patients aged 75–79 years. When analysed by model 1, Adjuvant! Online overestimated 10-year cumulative recurrence by 8·7% ([6·7–10·7], p<0·0001) overall, as well as in almost all subgroups. The appendix shows additional analyses with model 2 in which we used the comorbidity classification by the expert panel. The comorbidity classification that was defined by the expert panel was significantly associated with overall survival for “minor problems” (HR 1·2 [95% CI 0·7–1·9]; p=0·579), for “major problems+10” (HR 1·0 [0·8–1·2]; p=0·819), “major problems+20” (HR 1·3 [1·0–1·6]; p=0·036) and for “major problems+30” (HR 2·0 [1·6–2·5]; p<0·0001), when compared with “perfect health” (overall p<0·0001). By contrast with model 1, when using model 2, Adjuvant! Online underestimated the 10-year overall survival by –17·1% ([95% CI –21·0 to –13·2], p<0·0001, appendix) but Adjuvant! Online predicted 10-year cumulative recurrence accurately in all patients (–0·7% [95% CI –2·7–1·3], p=0·48). However, the predicted and observed cumulative recurrence especially differed in patients with several comorbidities (both according to the comorbidity status, and the absolute number of comorbidities). Figure 1 shows calibration plots of models 1 and 2 for overall survival and cumulative recurrence. For both overall survival and cumulative recurrence, observed and predicted outcomes differed significantly from the ideal relation between observed and predicted outcomes for both models (p<0·0001 for both models, figure 1). For overall survival, the c-index was 0·75 (95% CI 0·72–0·77) for model 1 and 0·70 (95% CI 0·68–0·73) for model 2 (figure 2). To investigate whether Adjuvant! Online predicted overall survival better than age at diagnosis, we additionally calculated the ROC for calendar age. For age, the c-index was 0·70 (95% CI 0·68–0·73). For cumulative recurrence, the c-index was 0·67 (95% CI 0·65–0·71) for model 1 and 0·62 (0·59–0·66) for model 2. Again, we calculated the ROC for calendar age, corresponding with a c-index of 0·48 (0·45–0·52).
0·4
Discussion Our findings show that Adjuvant! Online does not accurately predict overall survival and disease recurrence in older patients with early breast cancer. When comparing the discriminatory accuracy with cutoff points that were used in previous studies, we concluded that the discrimination of Adjuvant! Online according to the ROC was moderate for overall survival but poor for cumulative recurrence, whereas calibration was poor for both overall survival as well as recurrence.24,26
0·2
C-statistics Model 1: 0·67 (95% CI 0·65–0·71) Model 2: 0·62 (95% CI 0·59–0·66) Age: 0·48 (95% CI 0·45–0·52)
0 0
0·2
0·4 0·6 1-specificity
0·8
1·0
Figure 2: Receiver operating characteristic curves for 10-year observed versus predicted overall survival (A) and predicted cumulative recurrence (B)
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
5
Articles
Panel: Research in context Systematic review A recent systematic review was done by Engelhardt and colleagues, one of the co-authors on this paper.11 They searched PubMed, EMBASE, and Web of Science for all validation studies of Adjuvant! Online in breast cancer that were published in English before July, 2012, using the keywords “breast cancer”’; “adjuvant! Online” and “adjuvant therapy”. 11 The review identified six studies that assessed the validity of Adjuvant! Online. Conclusions from the review as well as the key findings of individual validation studies are discussed in our paper. Interpretation To our knowledge this is the first study to assess the validity of Adjuvant! Online in older patients with breast cancer with sufficient power. This study shows that Adjuvant! Online does not accurately predict survival and recurrence in older patients with breast cancer. We suggest that Adjuvant! Online’s predictions for older patients should be interpreted with caution.
Therefore, those results might not be representative for the general older breast cancer population. By contrast, our study included unselected older patients from a population-based cohort, which could explain the discrepancy between the studies. Several underlying mechanisms could explain our findings. First, the results could be explained by the fact that only patients aged 69 years or younger were included in the original study population from which Adjuvant! Online was developed.4 It is known that older patients differ from younger patients in many aspects.13,27 Consequently, results obtained in younger patients might not necessarily apply to older patients. Second, the predicted benefits of adjuvant systemic therapy itself were based on clinical trials that included relatively few older patients.1 Several studies have shown that older patients who are included in clinical trials are usually healthier than the general older population because of restrictions based on comorbidity status and functional status.28,29 Consequently, the absolute benefit of adjuvant treatment is likely to be less pronounced in the general population compared with a trial population. It is essential to distinguish relative and absolute benefits of treatment in older patients, since the increasing risk of competing mortality in older patients might lead to smaller absolute benefits of treatment. Therefore, findings from clinical trials might not always be extrapolated to the general older breast cancer population,29 and could explain part of the poor predictions of Adjuvant! Online in our study. However, there are promising ongoing trials which are specifically designed for older patients. For example, the Aster 70s trial (EudraCT 2011–00474422) investigates the effect of chemotherapy on overall 6
survival in patients aged 70 years and older with ERpositive breast cancer, who are selected with a new prediction method that takes competing risk into account.30 We noted substantial differences in outcomes for the two models used to enter comorbidity in Adjuvant! Online. The model in which comorbidity of all patients was defined as average for age overestimated both overall survival and cumulative recurrence. By contrast, the model in which comorbidity level was identified by the expert panel, and therefore expected to be the most accurate, strongly underestimated the actual 10-year overall survival, whereas it both underestimated and overestimated cumulative recurrence in different subgroups. The large difference in predicted overall survival of the two models can be explained by the way that comorbidity is incorporated in predictions made by Adjuvant! Online. In our specific comorbidity classification, most patients (80%) were placed in one of the three “major problems” categories, which is more serious than the “average for age” category. Therefore, the predicted overall survival in model 2 (using the comorbidity classification), will always be lower than when using model 1 (using “average for age” for all patients). Additionally, the observed differences between subgroups in both models are most likely explained by the small number of patients receiving chemotherapy or the combination of hormonal therapy and chemotherapy, as shown by the broad 95% CIs. As Adjuvant! Online does not provide a definition of the categories of comorbidity, there will be variation in how clinicians can define and enter comorbidity, potentially leading to inaccurate estimates of overall survival and recurrence. There are other risk prediction programs besides Adjuvant! Online. A recent systematic review identified 20 risk prediction models, of which most consisted of genetic risk prediction models.11 Of all these models, Adjuvant! Online is the only model that incorporates comorbidity, and is the most widely used and well known prediction tool.11 Furthermore, the prognostic accuracy of these other models is still unclear since they have not been developed nor sufficiently validated for older patients with breast cancer.11 Therefore, we propose that an improved prediction model specifically for older patients should be developed to individualise clinical decision making and improve outcomes in this heterogeneous and growing population. To our knowledge, this is the first study that has investigated the accuracy of Adjuvant! Online in a large and population-based cohort of older breast cancer patients (panel). The main strength of the analyses is the use of the FOCUS cohort, since it contains detailed information of a large number of unselected patients. Moreover, detailed information on comorbidity and geriatric variables enabled us to assess a comorbidity classification that was most likely to portray daily
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
practice. This study also has some limitations. We were unable to assess breast cancer-specific survival since information on cause of death was not available for all patients. However, patients with non-metastasised breast cancer who underwent primary surgical treatment are unlikely to die of breast cancer without developing distant metastases. Additionally, cause of death extracted from death certificates of patients with cancer is not always accurate.31,32 This issue is of great importance in older patients, since the risk of competing potential causes of mortality substantially increases with age.2 Therefore, in our opinion recurrence is a valid alternative for breast cancer-specific survival in older patients with breast cancer. Another limitation is that the administration of systemic therapy might have affected the results, as, with increasing age, patients are less often treated according to treatment guidelines.33 Therefore, administration of treatment might be associated with general health status and thereby affect the outcomes under study. While we are unable to disentangle the effect of treatment on outcome, the difference between observed and predicted outcomes was similar for all treatment subgroups. Finally, followup time for recurrence was not complete, which might have led to an underestimation of 10-year cumulative recurrence. However, the largest proportion of breast cancer recurrence occurs within 5 years of breast cancer diagnosis, a trend which our data followed.1 The finding that Adjuvant! Online predictions did not essentially differ from predictions based on calendar age, implies that the model might be of limited value in older patients with breast cancer. Our results therefore suggest that Adjuvant! Online predictions in patients with breast cancer aged 65 years or older need to be interpreted with caution. We propose that a prediction method specific to the older patients with breast cancer should be developed. Tumour characteristics as well as detailed and standardised patient characteristics should be included in the model to take in to account the large heterogeneity in phenotypes of the older population with breast cancer, thereby individualising clinical decision making and optimising outcome in older patients with breast cancer. Contributors NdG, WvdW, EE, EB, AdC, JP, JK, and GJL contributed to study design. NdG, WvdW, and HP performed data analyses. NdG provided the figures. All authors were responsible for data interpretation and writing of the report. Declaration of interests We declare no competing interests. Acknowledgments This work was supported by the Dutch Cancer Foundation (KWF 2007-3968). The authors would like to thank the Comprehensive Cancer Centre Netherlands (Leiden region), all participating hospitals, and H D M Murk Jansen and D Buis for data collection and data entry. References 1 Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 2005; 365: 1687–17.
2
3
4
5
6
7
8
9
10
11
12 13
14
15
16
17 18
19
20 21
22
23 24
van de Water W, Markopoulos C, van de Velde CJ, et al. Association between age at diagnosis and disease-specific mortality among postmenopausal women with hormone receptor-positive breast cancer. JAMA 2012; 307: 590–97. Berry SD, Ngo L, Samelson EJ, Kiel DP. Competing risk of death: an important consideration in studies of older adults. J Am Geriatr Soc 2010; 58: 783–87. Ravdin PM, Siminoff LA, Davis GJ, et al. Computer program to assist in making decisions about adjuvant therapy for women with early breast cancer. J Clin Oncol 2001; 19: 980–91. Olivotto IA, Bajdik CD, Ravdin PM, et al. Population-based validation of the prognostic model ADJUVANT! for early breast cancer. J Clin Oncol 2005; 23: 2716–25. Mook S, Schmidt MK, Rutgers EJ, et al. Calibration and discriminatory accuracy of prognosis calculation for breast cancer with the online Adjuvant! program: a hospital-based retrospective cohort study. Lancet Oncol 2009; 10: 1070–76. Hajage D, de Rycke Y, Bollet M, et al. External validation of Adjuvant! Online breast cancer prognosis tool. Prioritising recommendations for improvement. PLoS One 2011; 6: e27446. Campbell HE, Taylor MA, Harris AL, Gray AM. An investigation into the performance of the Adjuvant! Online prognostic programme in early breast cancer for a cohort of patients in the United Kingdom. Br J Cancer 2009; 101: 1074–84. Yao-Lung K, Dar-Ren C, Tsai-Wang C. Accuracy validation of adjuvant! online in Taiwanese breast cancer patients—a 10-year analysis. BMC Med Inform Decis Mak 2012; 12: 108. Bhoo-Pathy N, Yip CH, Hartman M, et al. Adjuvant! Online is overoptimistic in predicting survival of Asian breast cancer patients. Eur J Cancer 2012; 48: 982–89. Engelhardt EG, Garvelink MM, de Haes JH, et al. Predicting and Communicating the Risk of Recurrence and Death in Women With Early-Stage Breast Cancer: A Systematic Review of Risk Prediction Models. J Clin Oncol 2013; 32: 238–50. NABON. Richtlijn Mammacarcinoom versie 2.0. 13-2-2012. [in Dutch] (accessed April 10, 2014). Wildiers H, Kunkler I, Biganzoli L, et al. Management of breast cancer in elderly individuals: recommendations of the International Society of Geriatric Oncology. Lancet Oncol 2007; 8: 1101–15. Biganzoli L, Wildiers H, Oakman C, et al. Management of elderly patients with breast cancer: updated recommendations of the International Society of Geriatric Oncology (SIOG) and European Society of Breast Cancer Specialists (EUSOMA). Lancet Oncol 2012; 13: e148–160. Barnett K, Mercer SW, Norbury M, Watt G, Wyke S, Guthrie B. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380: 37–43. Patnaik JL, Byers T, Diguiseppi C, Denberg TD, Dabelea D. The influence of comorbidities on overall survival among older women diagnosed with breast cancer. J Natl Cancer Inst 2011; 103: 1101–11. Kendal WS. Dying with cancer: the influence of age, comorbidity, and cancer site. Cancer 2008; 112: 1354–62. de Glas NA, Kiderlen M, Bastiaannet E, et al. Postoperative complications and survival of elderly breast cancer patients: a FOCUS study analysis. Breast Cancer Res Treat 2013; 138: 561–69. Kiderlen M, de Glas NA, Bastiaannet E, et al. Diabetes in relation to breast cancer relapse and all-cause mortality in elderly breast cancer patients: a FOCUS study analysis. Ann Oncol 2013; 24: 3011–16. WHO. The ICD-10 classification of mental and behavioural disorders. Geneva: World Health Organization, 1992, 10: 1–267. Jager KJ, van Dijk PC, Zoccali C, Dekker FW. The analysis of survival data: the Kaplan-Meier method. Kidney Int 2008; 74: 560–65. Mell LK, Jeong JH, Nichols MA, Polite BN, Weichselbaum RR, Chmura SJ. Predictors of competing mortality in early breast cancer. Cancer 2010; 116: 5365–73. Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks and multi-state models. Stat Med 2007; 26: 2389–430. Crowson CS, Atkinson EJ, Therneau TM. Assessing calibration of prognostic risk scores. Stat Methods Med Res 2014; published online April 7. DOI:10.1177/0962280213497434.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
7
Articles
25 26 27
28
29
8
Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005; 61: 92–105. Akobeng AK. Understanding diagnostic tests 3: Receiver operating characteristic curves. Acta Paediatr 2007; 96: 644–47. Bastiaannet E, Liefers GJ, de Craen AJ, et al. Breast cancer in elderly compared to younger patients in the Netherlands: stage at diagnosis, treatment and survival in 127,805 unselected patients. Breast Cancer Res Treat 2010; 124: 801–07. Zulman DM, Sussman JB, Chen X, Cigolle CT, Blaum CS, Hayward RA. Examining the evidence: a systematic review of the inclusion and analysis of older adults in randomized controlled trials. J Gen Intern Med 2011; 26: 783–90. van de Water W, Kiderlen M, Bastiaannet E, et al. External validity of a trial comprised of elderly patients with hormone receptor-positive breast cancer. J Natl Cancer Inst 2014; 106: dju051.
30
31
32
33
Brain E, Baffert S, Bonnefoi H, et al. ASTER 70s: Benefit of adjuvant chemotherapy for oestrogen receptor-positive HER2negative breast carcinoma in women over 70 according to Genomic Grade. A French UNICANCER Geriatric Oncology Group (GERICO) and Breast Group (UCBG) multicentre phase III trial. Cancer Research 2012; 72: OT3-3–10. Hu CY, Xing Y, Cormier JN, Chang GJ. Assessing the utility of cancer-registry-processed cause of death in calculating cancer-specific survival. Cancer 2013; 119: 1900–07. Goldoni CA, Bonora K, Ciatto S, et al. Misclassification of breast cancer as cause of death in a service screening area. Cancer Causes Control 2009; 20: 533–38. van de Water W, Bastiaannet E, Dekkers OM, et al. Adherence to treatment guidelines and survival in patients with early-stage breast cancer by age at diagnosis. Br J Surg 2012; 99: 813–20.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Validity of Adjuvant! Online program in older patients with breast cancer: a population-based study Nienke A de Glas, Willemien van de Water, Ellen G Engelhardt, Esther Bastiaannet, Anton J M de Craen, Judith R Kroep, Hein Putter, Anne M Stiggelbout, Nir I Weijl, Cornelis J H van de Velde, Johanneke E A Portielje, Gerrit-Jan Liefers
Summary Background Adjuvant! Online is a prediction tool that can be used to aid clinical decision making in patients with breast cancer. It was developed in a patient population aged 69 years or younger, and subsequent validation studies included small numbers of older patients. Since older patients with breast cancer differ from younger patients in many aspects, the aim of this study was to investigate the validity of Adjuvant! Online in a large cohort of unselected older patients. Methods We included patients from the population-based FOCUS cohort, which included all consecutive patients aged 65 years or older who were diagnosed with invasive or in-situ breast cancer between Jan 1, 1997, and Dec 31, 2004, in the southwestern part of the Netherlands. We included all patients who fulfilled the criteria as stated by Adjuvant! Online: patients with unilateral, unicentric, invasive adenocarcinoma; no evidence of metastatic or residual disease; no evidence of T4 features; and no evidence of inflammatory breast cancer. We entered data from all patients with the “average for age” comorbidity status (model 1) and with an individualised comorbidity status (model 2). Findings We included 2012 patients. Median age of patients in the cohort was 74·0 years (IQR 69·0–79·0). 904 (45%) of 2012 patients died during follow-up, whereas 326 (16%) patients had recurrence. Median follow-up for overall survival was 9·0 years (IQR 7·4–10·7), and 6·6 years (4·4–6·6) for patients without recurrence. Using model 1, Adjuvant! Online overestimated 10-year overall survival by 9·8% ([95% CI 5·9–13·7], p<0·0001) and 10-year cumulative recurrence survival by 8·7% ([6·7–10·7], p<0·0001). By contrast, when using model 2, Adjuvant! Online underestimated the 10-year overall survival by –17·1% ([95% CI –21·0 to –13·2], p<0·0001). However, when using model 2, Adjuvant! Online predicted cumulative recurrence accurately in all patients (–0·7% [95% CI –2·7–1·3], p=0·48). Interpretation Adjuvant! Online does not accurately predict overall survival and recurrence in older patients with early breast cancer. Funding Dutch Cancer Foundation.
Introduction Adjuvant systemic treatment in early stage breast cancer is aimed at preventing breast cancer recurrence and mortality. The prognosis for early stage breast cancer has greatly improved with the introduction of adjuvant treatments.1 Optimum combinations of adjuvant endocrine therapy and chemotherapy now result in relative risk reductions of between 20% to 57% in 15-year mortality.1 In patients with a high absolute risk of recurrence, the potential benefit of adjuvant treatment is large. However, the benefit of adjuvant therapy might be attenuated in the presence of increased comorbidities or old age because of shorter life expectancy and competing causes of death.1–3 Therefore, the correct identification of patients who have an increased absolute risk of breast cancer recurrence and breast cancer mortality, and who might benefit from adjuvant therapy, is essential. In clinical practice, many prediction methods are used to estimate the benefit of adjuvant treatment. Of these prediction methods Adjuvant! Online is best known.4 Adjuvant! Online is an online, open-access prediction program that predicts 10-year breast cancer recurrence, breast cancer mortality, mortality due to other causes,
and expected benefits of specific adjuvant treatment options for individual patients.4 Predictions are based on six clinical factors, including comorbidity. The model was developed using a large database that was derived from the Surveillance, Epidemiology, End-results (SEER) registry, with a population of 34 252 patients.4 The study population consisted of women aged 35–69 years, who were diagnosed with early breast cancer between 1988 and 1992 in the USA.5 Adjuvant! Online has been validated in Canada, the USA, and in several Asian and European studies.5–11 Although Adjuvant! Online was developed in a cohort with quite young patients, the Dutch national guideline recommends using Adjuvant! Online for decision making in older patients (those aged 70 years or older) with breast cancer, on the basis of the original Canadian validation study.5,12 Treatment recommendations of the International Society of Geriatric Oncology from 2007 included using Adjuvant! Online in clinical decision making, although the most recent update did not mention the use of Adjuvant! Online.13,14 However, older patients make up a heterogeneous group, partly because of the presence of additional comorbidities.15,16 Survival is
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Lancet Oncol 2014 Published Online May 14, 2014 http://dx.doi.org/10.1016/ S1470-2045(14)70200-1 Department of Surgery (N A de Glas MD, W van de Water MD, E Bastiaannet PhD, Prof C J H van de Velde PhD, G-J Liefers PhD), Department of Gerontology and Geriatrics (N A de Glas, W van de Water, E Bastiaannet, A J M de Craen PhD), Department of Medical Decision Making (E G Engelhardt MSc, Prof A M Stiggelbout PhD), Department of Medical Oncology (J R Kroep PhD), Department of Medical Statistics (Prof Hein Putter PhD), Leiden University Medical Centre, Leiden, Netherlands; Department of Medical Oncology, Bronovo Hospital The Hague, The Hague, Netherlands (N I Weijl PhD); and Department of Medical Oncology, Haga Hospital The Hague, Leyweg The Hague, Netherlands (J E A Portielje PhD) Correspondence to: Dr Gerrit-Jan Liefers, Leiden University Medical Centre, Department of Surgery, Postzone K6-R, PO Box 9600, 2300 RC Leiden, Netherlands [email protected]
1
Articles
shorter in older patients with breast cancer compared with younger patients with breast cancer, mainly because the risk of competing mortality increases with age.3,17 Furthermore, older patients were under-represented in the trials which were used to predict the benefits of adjuvant therapy.1 This might hamper the generalisability of trial results to older patients with breast cancer. These factors probably affect predictions made by Adjuvant! Online. Additionally, results from a recent systematic review showed that the validity of its estimates has not been assessed with sufficient power in older patients.11 The aim of this study was to investigate the validity of Adjuvant! Online by assessing its discriminatory accuracy and calibration in a large cohort of women aged 65 years or older diagnosed with in-situ and invasive breast cancer.
Methods Patients and study design We used data from the population-based FOCUS cohort.18,19 This cohort includes all patients with incident breast cancer aged 65 years or older who were diagnosed in the geographically defined Comprehensive Cancer Center Region West in The Netherlands between Jan 1, 1997, and Dec 31, 2004 (n=3672). There were no exclusion criteria. Patients were identified through the Netherlands cancer registry, which registers information about all patients with cancer in the Netherlands through the central pathology database. Trained personnel reviewed the charts of all identified patients, and obtained information on specific treatments, comorbidity according to the ICD-10 classification,20 adverse events, geriatric variables (difficulty walking, sensory handicap, dementia, polypharmacy, and nursing home resident), and recurrences. Follow-up mortality data was available until Jan 1, 2011, by linking cancer registry data with municipal population registries. In the present study, all patients who fulfilled the criteria as stated by Adjuvant! Online were included: patients with unilateral, unicentric, invasive adenocarcinoma; no evidence of metastatic or residual disease; no evidence of T4 features; and no evidence of inflammatory breast cancer. Furthermore, patients had to have had previous definitive primary breast surgery and axillary node staging; not received neoadjuvant systemic treatment or radiotherapy; and be treated with radiotherapy if the patient had a lumpectomy.
Procedures We entered patient and tumour characteristics in the Adjuvant! Online program version 8·0, to calculate predicted 10-year overall survival and 10-year cumulative recurrence for every patient. We entered age as a continuous variable. In Adjuvant! Online, comorbidity is classified as “perfect health”; “minor problems”; “average for age”; “major problems+10”; “major problems+20”; and “major problems+30”, although cutoff points or 2
definitions of these different groups are not provided. Therefore, we used two different classifications of comorbidity. First, we entered the “average for age” category for all patients, as we expected that for all patients together, this category will give average outcomes of the predicted recurrence and survival rates that are most likely to portray the general population (model 1). Second, we assembled a panel consisting of one medical oncologist, one dedicated geriatric oncologist, one surgical oncologist, and three epidemiologists. The panel specified the categories of comorbidity on the basis of the effect of specific conditions on mortality, daily functioning, and quality of life (model 2). Patients were classified on the basis of the worst diagnosed condition. All analyses were done for both models. In Adjuvant! Online, pathological tumour size is classified as 0–1·0 cm, 1·1–2·0 cm, 2·1–3·0 cm, 3·1–5·0 cm, or larger than 5·0 cm. Exact tumour size was missing in 86 (4%) patients from our data (n=2012); for these patients, we used pathological T-stage instead. Since the categories of tumour size of Adjuvant! Online do not exactly resemble TNM-stage, we classified these 86 patients in the category that mostly resembled the T-stage as follows. Tumour size of patients with stage T1 was recoded as 1·1–2·0 cm, stage T2 was recoded as 3·1–5·0 cm, and stage T3 was recoded as larger than 5·0 cm. In Adjuvant! Online, tumour grade is defined as grade I, II, or III, or “undefined” in case of missing values. Nodal stage is classified as zero positive nodes, one to three positive nodes, four to nine positive nodes, or more than nine positive nodes. The exact number of positive nodes was missing in 298 (15%) of 2012 patients, as this information was often not registered in the pathology report. These patients with missing data for the number of positive axillary lymph nodes were classified using pathological N-stage (if available) or clinical N-stage as follows: N0 was recoded as no positive nodes, stage N1 was recoded as one to three positive nodes, stage N2 was recoded as four to nine positive nodes, and stage N3 was recoded as more than nine positive nodes. Finally, oestrogen receptor (ER) status was defined as positive, negative, or “undefined” in case of missing values. We used the definition of ER positivity that was given by the pathologist who rated the tumour sample. We assumed that missing data about both tumour grade and ER-status were completely missing at random. Whether patients received adjuvant endocrine therapy or chemotherapy was registered, but individual information on the specific treatment regimen was not available. Therefore, we defined the type of adjuvant therapy in line with the most commonly used treatments at the time of diagnosis. We defined endocrine therapy, if given, as tamoxifen only for patients who were diagnosed between 1997 and 1999 and as sequential therapy for patients who were diagnosed between 2000 and 2004. We
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
defined chemotherapy, if given, as first generation therapy for all patients.
Statistical analysis Primary outcome measures were 10-year overall survival and 10-year cumulative recurrence, which was defined as
n (%)
locoregional recurrence, distant recurrence, or contralateral breast cancer, in line with the definition of Adjuvant! Online. We first assessed the association between our comorbidity classification and overall survival to establish the validity of the classification using a Cox regression model. The model was adjusted for age only, since age was
10-year overall survival
10-year cumulative recurrence
Adjuvant predicted
Observed (SE)
Predicted–observed (95% CI)
2012 (100%)
48·8
39·0 (2)
65–69
560 (28%)
66·9
70–74
544 (27%)
57·9
75–79
417 (21%)
80–84
322 (16%)
≥85
169 (8%)
All patients
p value
Adjuvant predicted
Observed (SE)
Predicted–observed (95% CI)
p value
9·8 (5·9 to 13·7)
<0·0001
26·9
18·2 (1·0)
8·7 (6·7 to 10·7)
61·0 (3)
5·9 (0·01 to 12·8)
0·05
27·1
20·1 (2·0)
7·0 (3·1 to 10·9)
0·0005
51·0 (4)
6·9 (–1·0 to 14·8)
0·08
28·0
17·1 (1·8)
10·9 (7·4 to 14·4)
<0·0001
44·4
26·0 (3)
18·4 (12·5 to 24·3)
<0·0001
27·6
20·4 (2·2)
7·2 (2·9 to 11·5)
0·001
27·1
16·0 (4)
11·1 (3·2 to 19·0)
0·006
27·1
17·8 (2·4)
9·3 (4·6 to 14·0)
0·0001
11·3
5·0 (3)
6·3 (0·4 to 12·2)
0·04
20·8
9·1 (2·3)
11·7 (7·2 to 16·2)
<0·0001
<0·0001
Age (years)
Number of comorbidities 0
485 (24%)
52·2
55·0 (3)
–2·8 (–8·7 to 3·1)
0·35
27·4
19·7 (2·1)
7·7 (3·6 to 11·8)
0·0002
1
476 (24%)
50·9
50·0 (4)
0·9 (–7·0 to 8·8)
0·82
28·0
17·3 (2·0)
10·1 (6·8 to 14·6)
<0·0001
2
398 (20%)
47·3
33·0 (4)
14·3 (6·4 to 22·2)
<0·0001
26·2
17·1 (2·0)
9·1 (5·2 to 13·0)
0·0008
3
275 (14%)
47·4
31·0 (5)
16·4 (6·6 to 27·2)
0·001
25·9
17·1 (2·7)
8·8 (3·5 to 14·1)
0·001
≥4
378 (19%)
44·5
19·0 (3)
25·5 (19·6 to 31·4)
<0·0001
26·6
19·3 (2·2)
7·3 (3·0 to 11·6)
0·0009
Oestrogen receptor status Positive
1404 (79%)
42·8
37·0 (2)
5·8 (1·9 to 9·7)
0·004
24·4
17·7 (1·2)
6·7 (4·3 to 9·1)
<0·0001
Negative
370 (29%)
49·9
39·0 (4)
10·9 (3·0 to 18·8)
0·006
36·1
22·9 (2·2)
13·2 (8·9 to 17·5)
<0·0001
Undefined*
238
51·0
43·0 (4)
8·0 (0·1 to 15·9)
0·05
27·3
13·9 (2·5)
13·4 (8·5 to 18·3)
<0·0001
Tumour grade Grade 1
293 (20%)
57·2
45·0 (5)
12·2 (2·4 to 22·0)
0·01
17·2
10·2 (3·5)
7·0 (0·1 to 13·9)
0·05
Grade 2
682 (58%)
50·2
39·0 (4)
11·2 (3·4 to 19·1)
0·005
24·1
15·2 (1·7)
8·9 (5·6 to 12·2)
<0·0001
Grade 3
459 (22%)
43·6
35·0 (4)
8·6 (0·7 to 16·5)
0·03
33·9
29·0 (2·5)
Undefined*
578
46·8
40·0 (3)
6·8 (0·9 to 12·7)
0·02
29·7
18·1 (1·7)
0·1–1·0
299 (15%)
65·1
56·0 (5)
9·1 (–0·7 to 18·9)
0·07
17·0
8·5 (2·3)
8·5 (4·0 to 13·0)
0·0002
1·1–2·0
810 (40%)
54·8
45·0 (3)
9·8 (3·9 to 15·7)
0·001
21·2
14·9 (1·4)
6·3 (3·6 to 9·0)
<0·0001
2·1–3·0
540 (27%)
40·4
29·0 (3)
11·4 (5·5 to 17·3)
0·0001
32·3
23·6 (2·0)
8·7 (4·8 to 12·6)
<0·0001
3·1–5·0
301 (15%)
35·1
32·0 (4)
3·1 (–4·8 to 11·0)
0·44
37·4
24·9 (2·8)
12·5 (7·0 to 18·0)
<0·0001
62 (3%)
29·7
21·0 (7)
8·7 (–5·3 to 22·7)
0·21
51·4
28·6 (5·9)
22·8 (11·0 to 34·6)
0·0001
4·9 (–0·01 to 9·8) 11·6 (8·3 to 14·9)
0·05 <0·0001
Tumour size (cm)
>5 Positive nodes 0
1385 (69%)
53·9
43·0 (2)
10·9 (7·0 to 14·8)
<0·0001
21·8
13·6 (1·1)
8·2 (6·0 to 10·4)
<0·0001
1–3
475 (24%)
40·4
34·0 (3)
6·4 (0·5 to 12·3)
0·03
33·3
24·4 (2·3)
8·9 (4·4 to 13·4)
0·0001
4–9
100 (5%)
31·2
18·0 (5)
13·2 (3·3 to 23·1)
0·008
47·0
36·7 (5·0)
10·3 (0·4 to 20·2)
0·04
>9
52 (3%)
21·2
24·0 (9)
–2·8 (–20·9 to 15·3)
0·76
66·5
44·0 (7·5)
22·5 (7·4 to 37·6)
0·003
781 (39%)
60·4
55·0 (3)
5·4 (–0·5 to 11·3)
0·07
23·8
14·6 (1·4)
9·2 (6·5 to 12·0)
<0·0001
1231 (61%)
41·4
29·0 (2)
12·4 (8·5 to 16·3)
<0·0001
28·9
20·5 (1·3)
8·4 (5·9 to 11·0)
<0·0001
1133 (56%)
51·6
40·0 (2)
11·6 (7·7 to 15·5)
<0·0001
24·9
13·4 (1·1)
11·5 (9·3 to 13·7)
<0·0001
778 (39%)
44·0
38·0 (3)
6·0 (0·1 to 11·9)
0·05
28·9
23·4 (1·8)
5·5 (2·0 to 9·0)
0·002
Chemotherapy only
52 (3%)
52·2
35·0 (13)
17·2 (–8·9 to 43·3)
0·19
36·9
18·3 (5·6)
18·6 (7·4 to 29·8)
0·001
Hormonal therapy and chemotherapy
49 (2%)
53·2
43·0 (11)
10·2 (–11·9 to 32·3)
0·36
32·5
56·7 (11·5)
24·2 (–1·1 to 47·3)
0·04
Most extensive surgery type Breast conserving surgery Mastectomy Systemic treatment None Hormonal therapy only
ER=oestrogen receptor. *The category “undefined” was not included in the given percentages.
Table: Baseline characteristics and predicted versus observed 10 year overall survival and cumulative recurrence (model 1)
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
3
Articles
likely to be the only confounder in the relation between the comorbidity classification and overall survival. Second, we calculated observed 10-year overall survival using life tables derived from the Kaplan-Meier method.21 We calculated observed 10-year cumulative recurrence using cumulative incidence rates of breast cancer recurrence. Cause-specific outcomes might be affected by the risk of competing endpoints, particularly in older populations.22 Therefore, we estimated cumulative breast cancer recurrence by regression analyses according to Fine and Gray to assess the risk of a breast cancer recurrence while taking into account the risks of reaching other, competing endpoints (death without recurrence).23
See Online for appendix
A Model 1 Model 2 Perfect line (x=y)
100
p<0·0001
Observed 10-year overall survival (%)
80
60 p<0·0001 40
20
We compared the observed and predicted 10-year overall survival and cumulative recurrence using onesample Z-tests, with the predicted survival as fixed value and the observed value as the assessed variable. Subgroups were based on the information requested by Adjuvant! Online to predict outcomes as described here. Additionally, we calculated observed and predicted outcomes for subgroups based on the number of comorbidities (0, 1, 2, 3, and 4 or more), most extensive surgery (breast conserving surgery and mastectomy), and systemic treatment (none, endocrine therapy only, chemotherapy only, endocrine therapy, and chemotherapy). Furthermore, we assessed the calibration of the model by plotting the difference between observed and predicted outcomes for overall survival and cumulative recurrence. For this, we grouped patients into 10% intervals of the predicted 10-year overall survival. Intervals with fewer than 100 patients were combined. We tested regression lines and corresponding p values of the calibration plot against the so-called ideal line (x=y) using Poisson regression models.24 Additionally, we assessed the discriminatory accuracy by composing receiver-operator curves (ROC) and corresponding c-indices. C-indices are derived by calculating the area under the curve (AUC) of the ROC. A c-index of 1·0 means that the model is perfectly accurate, whereas a c-index of 0·50 means that the model does not predict better than chance.25 To investigate the added value of the prediction made by Adjuvant! Online, we additionally computed c-indices that were derived by using age as sole predictor. We did all analyses in SPSS version 20.0 and R version 2.15.3.
0 0
20
40
60
80
100
Adjuvant! Online predicted 10-year overall survival (%)
The FOCUS project and this analysis were funded by the Dutch Cancer Foundation (KWF 2007-3968). The Dutch Cancer Foundation was not involved in study design, collection, analysis, interpretation of the data, writing the report, nor the decision to submit the paper for publication. The corresponding author had full access to all data and had final responsibility for the decision to submit for publication.
B
Observed 10-year cumulative recurrence (%)
100
80
60
Results p<0·0001
40
p<0·0001 20
0 0
20
40
60
80
Adjuvant! online predicted 10-year cumulative recurrence (%)
Figure 1: Observed versus predicted 10-year overall survival (A) and cumulative recurrence (B)
4
Role of the funding source
100
We included 2012 patients who fulfilled the criteria as stated by Adjuvant! Online (appendix). Median age was 74·0 years (IQR 69·0–79·0), and 1527 (76%) of 2012 patients had one or more comorbidities (table). 904 (45%) of 2012 patients died during follow-up, and 326 (16%) patients developed a recurrence. Median follow-up was 9·0 years (IQR 7·4–10·7) for overall survival, and 6·6 years (IQR 4·4–6·6) for patients without disease recurrence. Of all 896 patients with complete 10-year follow-up, 177 (20%) patients had a recurrence, of which 131 (74%) occurred within the first 5 years.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
These findings contrast with previous validation studies that were done in younger patients with breast cancer. Most studies included very few patients older than 65 years of age,7–10 but some did report age-specific differences.5,6 The initial validation study concluded that breast cancer-specific survival was not accurately predicted in patients aged 76 years and older,5 which is confirmed in the current study. A European validation study including 817 patients aged 70 years and older concluded that although overall survival was not accurately predicted by Adjuvant! Online in patients aged 70 years and older, breast cancer-specific survival was.6 However, sample sizes were small compared with those in the present study and the authors used data from a selected group of patients from a tertiary centre. A
1·0
Model 1 Model 2 Age Reference line
0·8
Sensitivity
0·6
0·4
0·2
C-statistics Model 1: 0·75 (95% CI 0·72–0·77) Model 2: 0·70 (95% CI 0·68–0·73) Age: 0·70 (95% CI 0·68–0·73)
0
B
1·0
0·8
0·6 Sensitivity
The table shows both predicted and observed 10-year overall survival and 10-year cumulative recurrence according to model 1. Using model 1, Adjuvant! Online overestimated the 10-year overall survival by 9·8% ([95% CI 5·9–13·7], p<0·0001). The difference between observed and predicted 10-year overall survival was smallest in patients aged 65–69 years, and largest in patients aged 75–79 years. When analysed by model 1, Adjuvant! Online overestimated 10-year cumulative recurrence by 8·7% ([6·7–10·7], p<0·0001) overall, as well as in almost all subgroups. The appendix shows additional analyses with model 2 in which we used the comorbidity classification by the expert panel. The comorbidity classification that was defined by the expert panel was significantly associated with overall survival for “minor problems” (HR 1·2 [95% CI 0·7–1·9]; p=0·579), for “major problems+10” (HR 1·0 [0·8–1·2]; p=0·819), “major problems+20” (HR 1·3 [1·0–1·6]; p=0·036) and for “major problems+30” (HR 2·0 [1·6–2·5]; p<0·0001), when compared with “perfect health” (overall p<0·0001). By contrast with model 1, when using model 2, Adjuvant! Online underestimated the 10-year overall survival by –17·1% ([95% CI –21·0 to –13·2], p<0·0001, appendix) but Adjuvant! Online predicted 10-year cumulative recurrence accurately in all patients (–0·7% [95% CI –2·7–1·3], p=0·48). However, the predicted and observed cumulative recurrence especially differed in patients with several comorbidities (both according to the comorbidity status, and the absolute number of comorbidities). Figure 1 shows calibration plots of models 1 and 2 for overall survival and cumulative recurrence. For both overall survival and cumulative recurrence, observed and predicted outcomes differed significantly from the ideal relation between observed and predicted outcomes for both models (p<0·0001 for both models, figure 1). For overall survival, the c-index was 0·75 (95% CI 0·72–0·77) for model 1 and 0·70 (95% CI 0·68–0·73) for model 2 (figure 2). To investigate whether Adjuvant! Online predicted overall survival better than age at diagnosis, we additionally calculated the ROC for calendar age. For age, the c-index was 0·70 (95% CI 0·68–0·73). For cumulative recurrence, the c-index was 0·67 (95% CI 0·65–0·71) for model 1 and 0·62 (0·59–0·66) for model 2. Again, we calculated the ROC for calendar age, corresponding with a c-index of 0·48 (0·45–0·52).
0·4
Discussion Our findings show that Adjuvant! Online does not accurately predict overall survival and disease recurrence in older patients with early breast cancer. When comparing the discriminatory accuracy with cutoff points that were used in previous studies, we concluded that the discrimination of Adjuvant! Online according to the ROC was moderate for overall survival but poor for cumulative recurrence, whereas calibration was poor for both overall survival as well as recurrence.24,26
0·2
C-statistics Model 1: 0·67 (95% CI 0·65–0·71) Model 2: 0·62 (95% CI 0·59–0·66) Age: 0·48 (95% CI 0·45–0·52)
0 0
0·2
0·4 0·6 1-specificity
0·8
1·0
Figure 2: Receiver operating characteristic curves for 10-year observed versus predicted overall survival (A) and predicted cumulative recurrence (B)
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
5
Articles
Panel: Research in context Systematic review A recent systematic review was done by Engelhardt and colleagues, one of the co-authors on this paper.11 They searched PubMed, EMBASE, and Web of Science for all validation studies of Adjuvant! Online in breast cancer that were published in English before July, 2012, using the keywords “breast cancer”’; “adjuvant! Online” and “adjuvant therapy”. 11 The review identified six studies that assessed the validity of Adjuvant! Online. Conclusions from the review as well as the key findings of individual validation studies are discussed in our paper. Interpretation To our knowledge this is the first study to assess the validity of Adjuvant! Online in older patients with breast cancer with sufficient power. This study shows that Adjuvant! Online does not accurately predict survival and recurrence in older patients with breast cancer. We suggest that Adjuvant! Online’s predictions for older patients should be interpreted with caution.
Therefore, those results might not be representative for the general older breast cancer population. By contrast, our study included unselected older patients from a population-based cohort, which could explain the discrepancy between the studies. Several underlying mechanisms could explain our findings. First, the results could be explained by the fact that only patients aged 69 years or younger were included in the original study population from which Adjuvant! Online was developed.4 It is known that older patients differ from younger patients in many aspects.13,27 Consequently, results obtained in younger patients might not necessarily apply to older patients. Second, the predicted benefits of adjuvant systemic therapy itself were based on clinical trials that included relatively few older patients.1 Several studies have shown that older patients who are included in clinical trials are usually healthier than the general older population because of restrictions based on comorbidity status and functional status.28,29 Consequently, the absolute benefit of adjuvant treatment is likely to be less pronounced in the general population compared with a trial population. It is essential to distinguish relative and absolute benefits of treatment in older patients, since the increasing risk of competing mortality in older patients might lead to smaller absolute benefits of treatment. Therefore, findings from clinical trials might not always be extrapolated to the general older breast cancer population,29 and could explain part of the poor predictions of Adjuvant! Online in our study. However, there are promising ongoing trials which are specifically designed for older patients. For example, the Aster 70s trial (EudraCT 2011–00474422) investigates the effect of chemotherapy on overall 6
survival in patients aged 70 years and older with ERpositive breast cancer, who are selected with a new prediction method that takes competing risk into account.30 We noted substantial differences in outcomes for the two models used to enter comorbidity in Adjuvant! Online. The model in which comorbidity of all patients was defined as average for age overestimated both overall survival and cumulative recurrence. By contrast, the model in which comorbidity level was identified by the expert panel, and therefore expected to be the most accurate, strongly underestimated the actual 10-year overall survival, whereas it both underestimated and overestimated cumulative recurrence in different subgroups. The large difference in predicted overall survival of the two models can be explained by the way that comorbidity is incorporated in predictions made by Adjuvant! Online. In our specific comorbidity classification, most patients (80%) were placed in one of the three “major problems” categories, which is more serious than the “average for age” category. Therefore, the predicted overall survival in model 2 (using the comorbidity classification), will always be lower than when using model 1 (using “average for age” for all patients). Additionally, the observed differences between subgroups in both models are most likely explained by the small number of patients receiving chemotherapy or the combination of hormonal therapy and chemotherapy, as shown by the broad 95% CIs. As Adjuvant! Online does not provide a definition of the categories of comorbidity, there will be variation in how clinicians can define and enter comorbidity, potentially leading to inaccurate estimates of overall survival and recurrence. There are other risk prediction programs besides Adjuvant! Online. A recent systematic review identified 20 risk prediction models, of which most consisted of genetic risk prediction models.11 Of all these models, Adjuvant! Online is the only model that incorporates comorbidity, and is the most widely used and well known prediction tool.11 Furthermore, the prognostic accuracy of these other models is still unclear since they have not been developed nor sufficiently validated for older patients with breast cancer.11 Therefore, we propose that an improved prediction model specifically for older patients should be developed to individualise clinical decision making and improve outcomes in this heterogeneous and growing population. To our knowledge, this is the first study that has investigated the accuracy of Adjuvant! Online in a large and population-based cohort of older breast cancer patients (panel). The main strength of the analyses is the use of the FOCUS cohort, since it contains detailed information of a large number of unselected patients. Moreover, detailed information on comorbidity and geriatric variables enabled us to assess a comorbidity classification that was most likely to portray daily
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
Articles
practice. This study also has some limitations. We were unable to assess breast cancer-specific survival since information on cause of death was not available for all patients. However, patients with non-metastasised breast cancer who underwent primary surgical treatment are unlikely to die of breast cancer without developing distant metastases. Additionally, cause of death extracted from death certificates of patients with cancer is not always accurate.31,32 This issue is of great importance in older patients, since the risk of competing potential causes of mortality substantially increases with age.2 Therefore, in our opinion recurrence is a valid alternative for breast cancer-specific survival in older patients with breast cancer. Another limitation is that the administration of systemic therapy might have affected the results, as, with increasing age, patients are less often treated according to treatment guidelines.33 Therefore, administration of treatment might be associated with general health status and thereby affect the outcomes under study. While we are unable to disentangle the effect of treatment on outcome, the difference between observed and predicted outcomes was similar for all treatment subgroups. Finally, followup time for recurrence was not complete, which might have led to an underestimation of 10-year cumulative recurrence. However, the largest proportion of breast cancer recurrence occurs within 5 years of breast cancer diagnosis, a trend which our data followed.1 The finding that Adjuvant! Online predictions did not essentially differ from predictions based on calendar age, implies that the model might be of limited value in older patients with breast cancer. Our results therefore suggest that Adjuvant! Online predictions in patients with breast cancer aged 65 years or older need to be interpreted with caution. We propose that a prediction method specific to the older patients with breast cancer should be developed. Tumour characteristics as well as detailed and standardised patient characteristics should be included in the model to take in to account the large heterogeneity in phenotypes of the older population with breast cancer, thereby individualising clinical decision making and optimising outcome in older patients with breast cancer. Contributors NdG, WvdW, EE, EB, AdC, JP, JK, and GJL contributed to study design. NdG, WvdW, and HP performed data analyses. NdG provided the figures. All authors were responsible for data interpretation and writing of the report. Declaration of interests We declare no competing interests. Acknowledgments This work was supported by the Dutch Cancer Foundation (KWF 2007-3968). The authors would like to thank the Comprehensive Cancer Centre Netherlands (Leiden region), all participating hospitals, and H D M Murk Jansen and D Buis for data collection and data entry. References 1 Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Effects of chemotherapy and hormonal therapy for early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet 2005; 365: 1687–17.
2
3
4
5
6
7
8
9
10
11
12 13
14
15
16
17 18
19
20 21
22
23 24
van de Water W, Markopoulos C, van de Velde CJ, et al. Association between age at diagnosis and disease-specific mortality among postmenopausal women with hormone receptor-positive breast cancer. JAMA 2012; 307: 590–97. Berry SD, Ngo L, Samelson EJ, Kiel DP. Competing risk of death: an important consideration in studies of older adults. J Am Geriatr Soc 2010; 58: 783–87. Ravdin PM, Siminoff LA, Davis GJ, et al. Computer program to assist in making decisions about adjuvant therapy for women with early breast cancer. J Clin Oncol 2001; 19: 980–91. Olivotto IA, Bajdik CD, Ravdin PM, et al. Population-based validation of the prognostic model ADJUVANT! for early breast cancer. J Clin Oncol 2005; 23: 2716–25. Mook S, Schmidt MK, Rutgers EJ, et al. Calibration and discriminatory accuracy of prognosis calculation for breast cancer with the online Adjuvant! program: a hospital-based retrospective cohort study. Lancet Oncol 2009; 10: 1070–76. Hajage D, de Rycke Y, Bollet M, et al. External validation of Adjuvant! Online breast cancer prognosis tool. Prioritising recommendations for improvement. PLoS One 2011; 6: e27446. Campbell HE, Taylor MA, Harris AL, Gray AM. An investigation into the performance of the Adjuvant! Online prognostic programme in early breast cancer for a cohort of patients in the United Kingdom. Br J Cancer 2009; 101: 1074–84. Yao-Lung K, Dar-Ren C, Tsai-Wang C. Accuracy validation of adjuvant! online in Taiwanese breast cancer patients—a 10-year analysis. BMC Med Inform Decis Mak 2012; 12: 108. Bhoo-Pathy N, Yip CH, Hartman M, et al. Adjuvant! Online is overoptimistic in predicting survival of Asian breast cancer patients. Eur J Cancer 2012; 48: 982–89. Engelhardt EG, Garvelink MM, de Haes JH, et al. Predicting and Communicating the Risk of Recurrence and Death in Women With Early-Stage Breast Cancer: A Systematic Review of Risk Prediction Models. J Clin Oncol 2013; 32: 238–50. NABON. Richtlijn Mammacarcinoom versie 2.0. 13-2-2012. [in Dutch] (accessed April 10, 2014). Wildiers H, Kunkler I, Biganzoli L, et al. Management of breast cancer in elderly individuals: recommendations of the International Society of Geriatric Oncology. Lancet Oncol 2007; 8: 1101–15. Biganzoli L, Wildiers H, Oakman C, et al. Management of elderly patients with breast cancer: updated recommendations of the International Society of Geriatric Oncology (SIOG) and European Society of Breast Cancer Specialists (EUSOMA). Lancet Oncol 2012; 13: e148–160. Barnett K, Mercer SW, Norbury M, Watt G, Wyke S, Guthrie B. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380: 37–43. Patnaik JL, Byers T, Diguiseppi C, Denberg TD, Dabelea D. The influence of comorbidities on overall survival among older women diagnosed with breast cancer. J Natl Cancer Inst 2011; 103: 1101–11. Kendal WS. Dying with cancer: the influence of age, comorbidity, and cancer site. Cancer 2008; 112: 1354–62. de Glas NA, Kiderlen M, Bastiaannet E, et al. Postoperative complications and survival of elderly breast cancer patients: a FOCUS study analysis. Breast Cancer Res Treat 2013; 138: 561–69. Kiderlen M, de Glas NA, Bastiaannet E, et al. Diabetes in relation to breast cancer relapse and all-cause mortality in elderly breast cancer patients: a FOCUS study analysis. Ann Oncol 2013; 24: 3011–16. WHO. The ICD-10 classification of mental and behavioural disorders. Geneva: World Health Organization, 1992, 10: 1–267. Jager KJ, van Dijk PC, Zoccali C, Dekker FW. The analysis of survival data: the Kaplan-Meier method. Kidney Int 2008; 74: 560–65. Mell LK, Jeong JH, Nichols MA, Polite BN, Weichselbaum RR, Chmura SJ. Predictors of competing mortality in early breast cancer. Cancer 2010; 116: 5365–73. Putter H, Fiocco M, Geskus RB. Tutorial in biostatistics: competing risks and multi-state models. Stat Med 2007; 26: 2389–430. Crowson CS, Atkinson EJ, Therneau TM. Assessing calibration of prognostic risk scores. Stat Methods Med Res 2014; published online April 7. DOI:10.1177/0962280213497434.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1
7
Articles
25 26 27
28
29
8
Heagerty PJ, Zheng Y. Survival model predictive accuracy and ROC curves. Biometrics 2005; 61: 92–105. Akobeng AK. Understanding diagnostic tests 3: Receiver operating characteristic curves. Acta Paediatr 2007; 96: 644–47. Bastiaannet E, Liefers GJ, de Craen AJ, et al. Breast cancer in elderly compared to younger patients in the Netherlands: stage at diagnosis, treatment and survival in 127,805 unselected patients. Breast Cancer Res Treat 2010; 124: 801–07. Zulman DM, Sussman JB, Chen X, Cigolle CT, Blaum CS, Hayward RA. Examining the evidence: a systematic review of the inclusion and analysis of older adults in randomized controlled trials. J Gen Intern Med 2011; 26: 783–90. van de Water W, Kiderlen M, Bastiaannet E, et al. External validity of a trial comprised of elderly patients with hormone receptor-positive breast cancer. J Natl Cancer Inst 2014; 106: dju051.
30
31
32
33
Brain E, Baffert S, Bonnefoi H, et al. ASTER 70s: Benefit of adjuvant chemotherapy for oestrogen receptor-positive HER2negative breast carcinoma in women over 70 according to Genomic Grade. A French UNICANCER Geriatric Oncology Group (GERICO) and Breast Group (UCBG) multicentre phase III trial. Cancer Research 2012; 72: OT3-3–10. Hu CY, Xing Y, Cormier JN, Chang GJ. Assessing the utility of cancer-registry-processed cause of death in calculating cancer-specific survival. Cancer 2013; 119: 1900–07. Goldoni CA, Bonora K, Ciatto S, et al. Misclassification of breast cancer as cause of death in a service screening area. Cancer Causes Control 2009; 20: 533–38. van de Water W, Bastiaannet E, Dekkers OM, et al. Adherence to treatment guidelines and survival in patients with early-stage breast cancer by age at diagnosis. Br J Surg 2012; 99: 813–20.
www.thelancet.com/oncology Published online May 14, 2014 http://dx.doi.org/10.1016/S1470-2045(14)70200-1