- Email: [email protected]
Partner status and survival after cancer: A competing risks analysis
Cancer Epidemiology 41 (2016) 16–23
Contents lists available at ScienceDirect
Cancer Epidemiology The International Journal of Cancer Epidemiology, Detection, and Prevention journal homepage: www.cancerepidemiology.net
Partner status and survival after cancer: A competing risks analysis Paramita Dasguptaa , Gavin Turrellb , Joanne F. Aitkena,b,c , Peter D. Baadea,b,d,* a
Cancer Council Queensland, P.O. Box 201, Spring Hill, QLD 4004, Australia School of Public Health and Social Work, Queensland University of Technology, Herston Road, Kelvin Grove, QLD 4059, Australia c School of Population Health, University of Queensland, Brisbane, Australia d Menzies Health Institute Queensland, Griffith University, Gold Coast Campus, Parklands Drive, Southport, QLD 4222, Australia b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 22 October 2015 Received in revised form 13 December 2015 Accepted 18 December 2015 Available online 14 January 2016
Objective: The survival benefits of having a partner for all cancers combined is well recognized, however its prognostic importance for individual cancer types, including competing mortality causes, is less clear. This study was undertaken to quantify the impact of partner status on survival due to cancer-specific and competing mortality causes. Methods: Data were obtained from the population-based Queensland Cancer Registry on 176,050 incident cases of ten leading cancers diagnosed in Queensland (Australia) from 1996 to 2012. Flexible parametric competing-risks models were used to estimate cause-specific hazards and cumulative probabilities of death, adjusting for age, stage (breast, colorectal and melanoma only) and stratifying by sex. Results: Both unpartnered males and females had higher total cumulative probability of death than their partnered counterparts for each site. For example, the survival disadvantage for unpartnered males ranged from 3% to 30% with higher mortality burden from both the primary cancer and competing mortality causes. The cause-specific age-adjusted hazard ratios were also consistent with patients without a partner having increased mortality risk although the specific effect varied by site, sex and cause of death. For all combined sites, unpartnered males had a 46%, 18% and 44% higher risk of cancer-specific, other cancer and non-cancer mortality respectively with similar patterns for females. The higher mortality risk persisted after adjustment for stage. Conclusions: It is important to better understand the mechanisms by which having a partner is beneficial following a cancer diagnosis, so that this can inform improvements in cancer management for all people with cancer. ã 2015 Elsevier Ltd. All rights reserved.
Keywords: Cancer Survival Partner status Inequalities Competing risks
1. Introduction The protective effect of marriage on overall cancer survival [1] has been well established. Emerging evidence also suggests that the survival benefit for married patients may be increasing over time [2]. The protective effect of marriage may reflect increased social support over the cancer continuum from diagnosis to survivorship to buffer the adverse effects of stress-related biological processes and immune responses [1,3]. Having a partner may also provide a source of financial, practical and emotional support while undergoing treatment and has been linked to more timely care, increased receipt of curative therapies, greater
* Corresponding author at: Cancer Council Queensland, P.O. Box 201, Spring Hill, QLD 4004, Australia. Fax: +61 7 3259 8527. E-mail addresses: [email protected] (P. Dasgupta), [email protected] (G. Turrell), [email protected] (J.F. Aitken), [email protected] (P.D. Baade). http://dx.doi.org/10.1016/j.canep.2015.12.009 1877-7821/ ã 2015 Elsevier Ltd. All rights reserved.
compliance with multi-modal treatments and improved psychosocial outcomes [1,3–9]. Results from studies that combine cancer sites may be affected by too much aggregation across different diseases. However findings to date on the impact of being married on cause-specific survival for individual cancer sites are equivocal with nonsignificant [6,10], protective [4,5,11–15] and mixed effects [16,17] being reported after adjusting for various combinations of known survival determinants. Some [4,12,16] but not all [15,18] studies also reported that the partner effect varied by gender. These inconsistencies likely reflect the heterogeneity in time period, cohort characteristic, covariates included in statistical models and definition of marital status across studies. While several categorized patients as married or unmarried based on official marital status categories [4,5,13,14] or extended the married group to also include partnered patients [11,12], others split the unmarried category into sub-groups (single, widowed, divorced or separated) [6,15–17].
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
The impact of partner status has typically being reported using net survival [4,10,12,15], describing the hypothetical situation where the diagnosed cancer is the only cause of death [19]. In the real world, however patients are at risk of death from various mutually exclusive causes. Hence partitioning overall mortality into different causes has the potential to provide a richer understanding of the impact having a partner has on survival after a cancer diagnosis [20,21]. To date, only a few studies have reported on the beneficial effects of having a partner on both cancer-specific [5,22–24] and competing mortality causes [22,24], however none have systematically compared these effects across multiple cancer sites. In this study we quantified the survival benefit of being partnered for each of the ten leading cancers diagnosed in Queensland, Australia using population-based data and competing risk methods [21]. 2. Methods Approval for the use of these de-identified data was obtained from the data custodian, Queensland Health. In addition, ethics and
17
data custodian approval for the extraction of stage information was obtained by The University of Queensland Social and Behavioural Sciences Ethical Review Committee and Queensland Health. Data were extracted from the population-based Queensland Cancer Registry, to which notifications of any cancer diagnosis (except keratinocyte cancers) is a statutory requirement [25]. All incident cases of the ten leading cancers diagnosed in Queensland from 1 January 1996 to 31st December 2012 (Table 1) were initially considered. Cases identified at death or autopsy, diagnosed with multiple primary cancers, with unknown marital status or who survived for less than one day were excluded. The cohort was restricted to patients aged 20–79 years at diagnosis consistent with an earlier published study [22]. While the Queensland Cancer Registry does not habitually gather information pertaining to stage at diagnosis, additional clinical information collected routinely [25] or manually extracted from pathology forms enabled a measure of stage at diagnosis to be calculated for melanoma [26], breast [27] and colorectal cancer [28,29]. To allow sufficient numbers for modelling, cases were collectively grouped as early (localized/locally advanced) or
Table 1 Demographic characteristics by partner status, sex and mortality outcomes for selected cancers, Queensland, 1996–2012. Site (Partner)
Males N
Females 10-year cumulative probability of death (%)
N
Totala
Cancer-specific
Other cancers
Non-cancer
10-year cumulative probability of death (%) Totala
Cancer-specific
Other cancers
Non-cancer
Colorectal No partner Has partner
4,654 12,930
59 [54,65] 51 [46,55]
41 [39,44] 36 [34,38]
3 [2,4] 4 [3,5]
15 [13,17] 11 [9,12]
5,649 7,413
53 [50,58] 46 [41,51]
40 [39,42] 36 [34,39]
3 [2,4] 2 [1,3]
10 [9,12] 8 [6,9]
Lung No partner Has partner
4,797 9,094
92 [88,97] 90 [86,94]
85 [83,87] 83 [81,85]
1 [0,2] 1 [0,2]
6 [5,8] 6 [5,7]
3,576 4,058
87 [82,93] 85 [80,91]
81 [78,84] 80 [77,83]
1 [0,2] 1 [0,2]
6 [4,7] 4 [3,6]
Melanoma No partner Has partner
3,221 9,454
44 [36,48] 35 [32,40]
19 [16,21] 14 [13,16]
8 [6,9] 8 [7,9]
17 [14,18] 13 [12,15]
3,222 5,922
30 [24,36] 22 [17,28]
11 [9,13] 9 [7,11]
6 [4,8] 4 [3,6]
13 [11,15] 9 [7,11]
Breast-female No partner Has partner
11,345 22,872
37 [33,41] 28 [24,31]
20 [19,22] 17 [15,18]
4 [3,5] 3 [2,4]
13 [11,14] 8 [7,9]
Uterus No partner Has partner
1,816 2,738
35 [26,43] 25 [18,31]
20 [16,23] 15 [12,17]
3 [1,5] 3 [1,5]
12 [9,15] 7 [5,9]
Prostate No partner Has partner
7,728 27,984
43 [38,49] 37 [33,41]
20 [18,23] 17 [15,19]
6 [5,7] 5 [4,6]
17 [15,19] 15 [14,16]
Kidney No partner Has partner
1,085 3,120
55 [45,66] 49 [40,57]
31 [27,36] 33 [29,37]
7 [4,9] 5 [3,7]
17 [14,21] 11 [8,13]
988 1,340
49 [38,63] 48 [40,55]
34 [27,40] 34 [30,36]
3 [2,7] 3 [2,5]
12 [9,16] 11 [8,14]
Bladder No partner Has partner
1,183 2,789
52 [42,63] 48 [41,57]
27 [23,32] 24 [21,28]
9 [6,11] 10 [8,12]
16 [13,20] 14 [12,17]
608 580
47 [31,62] 45 [36,59]
29 [21,36] 30 [26,38]
6 [2,9] 5 [3,7]
12 [8,17] 10 [7,14]
Non-Hodgkin Lymphoma No partner Has partner
1,182 3,499
55 [45,66] 45 [38,52]
36 [32,41] 29 [26,32]
5 [3,8] 6 [4,8]
14 [10,17] 10 [8,12]
1,447 2,202
42 [33,51] 40 [31,48]
28 [24,32] 28 [23,31]
4 [2,6] 4 [2,6]
10 [7,13] 8 [6,11]
Head and neck No partner Has partner
2,363 3,445
74 [65,83] 57 [48,65]
49 [45,53] 35 [31,38]
7 [5,9] 9 [6,11]
18 [15,21] 13 [11,16]
783 963
63 [46,78] 50 [36,65]
42 [34,49] 34 [28,41]
7 [3,10] 7 [3,11]
14 [9,19] 9 [5,13]
Overall (combined sites) No partner Has partner
26,213 72,315
58 [55,62] 46 [42,49]
39 [38,40] 30 [29,31]
5 [4,7] 5 [3,6]
14 [13,15] 11 [10,12]
29,434 48,088
48 [45,51] 40 [37,43]
33 [32,34] 29 [28,30]
4 [3,5] 3 [2,4]
11 [10,12] 8 [7,9]
a Sum of the three cause-specific probabilities of death. Cancer-specific are deaths due to primary cancer, other cancers are deaths due to any other cancers. Refer to text for further details.
18
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
advanced (regional, distant) for melanoma and colorectal cancer. For breast cancer, cases were defined as ‘early’ (Stage I) if 20 mm size with no evident nodal spread or metastases while Stage II, III and IV cancers which could not be distinguished based on available information were collectively categorized as ‘advanced’ [27]. For all three sites, cases with unknown stage were excluded. Information about marital status is collected routinely by the Queensland Cancer Registry, based on a patients self-reported data as noted in the medical record. For this study cases with a partner included those who were married or in a de facto relationship.
2.2. Statistical analysis All statistical analyses were performed with Stata/SE version 13 (StataCorp, TX, USA) and were stratified by sex due to differences in the age-specific patterns of partner status by sex [30].
2.1. Survival times The cohort was followed up to 31st December 2012 with matching to the Queensland Registrar of Births, Deaths and Marriages and the National Death Index [25] to identify
All: Partner
1.0 0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
2.2.1. Exploratory analysis Cohort characteristics were compared using the x2 test.
All: No partner
1.0
0.0
0.0 0
2
4
6
8
10
Colorectal: Partner
1.0
Probability of Death
Queensland-specific and interstate deaths respectively. Survival was measured in days from the date of diagnosis to death or the study end point, with cases alive at the study end point being censored. Cause of death information from the Queensland Cancer Registry [25] was categorized as cancer-specific if attributed to the primary cancer, other cancers or non-cancers. Follow-up was restricted to 10 years after diagnosis.
0
2
4
6
8
10
Colorectal: No partner
1.0
Lung: Partner
1.0
0.8
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
0
2
4
6
8
10
Melanoma: Partner
1.0
0
2
4
6
8
10
Melanoma: No partner
1.0
0.0 0
2
4
6
8
10
Prostate: Partner
1.0
0
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
2
4
6
8
10
Kidney: Partner
1.0
0
2
4
6
8
10
Kidney: No partner
1.0
2
4
6
8
10
Bladder: Partner
1.0
0
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
4
6
8
10
NHL: Partner
1.0
0
2
4
6
8
10
NHL: No partner
1.0
2
4
6
8
10
Head and neck: Partner
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
2
4
6
8
10
0
2
4
6
8
10
8
10
2
4
6
8
10
0.0 0
0.8
0
6
Bladder: No partner
1.0
0.6
2
4
0.0 0
0.8
0
2
Prostate: No partner
1.0
0.8
0
Lung: No partner
1.0
0
2
4
6
8
10
Head and neck: No partner
0.0 0
2
4
6
8
10
0
2
4
6
8
10
Time Since Diagnosis (Years) cancer−specific
other cancers
non−cancers
Fig 1. Stacked plots of cumulative probabilities of death by cancer site and partner status among males, Queensland 1996–2012. These plots show the cumulative probability of death due to cancer-specific, other cancers and non-cancers causes stacked on top of each other for each evaluated cancer site.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
2.2.2. Flexible parametric models Survival analyses were carried out within a flexible parametric modelling framework which use restricted cubic splines to model the baseline hazard function [31,32]. For each site, flexible parametric models on the log cumulative hazard scale were used to model the three competing mortality causes simultaneously [21,32]. Additional subgroup analysis by cancer stage were carried out for melanoma, colorectal and breast cancer. Models were adjusted for age, partner status and cancer stage where relevant. Baseline hazard functions were estimated separately for each cause by using the rcsbaseoff option during model fitting and including the three cause-specific indicators as main effects and time-dependent effects [32]. Interaction terms between each cause of death and age and partner status and stage were fitted. Time-dependent effects were assessed for all
All: Partner
1.0 0.8 0.6 0.4 0.2 0.0
Probability of Death
0
2
4
6
10
Colorectal: Partner
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
8
0
2
4
6
8
0
2
4
6
10
Bladder: Partner
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
0
0
10
Head and neck: Partner
1.0 0.8 0.6 0.4 0.2 0.0
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
8
10
6
8
10
2
4
6
8
2
4
6
8
2
4
6
8
0
2
4
6
8
Lung: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
0
10
0
2
0
2
6
8
10
4
6
8
4
6
8
0
2
4
6
0
10
0
0
10
6
8
10
2
4
6
8
10
2
4
6
8
10
NHL: No partner
1.0 0.8 0.6 0.4 0.2 0.0 8
4
Kidney: No partner
1.0 0.8 0.6 0.4 0.2 0.0 10
2
Breast: No partner
1.0 0.8 0.6 0.4 0.2 0.0
NHL: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
4
Kidney: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
2
Lung: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Breast: Partner
1.0 0.8 0.6 0.4 0.2 0.0
Bladder: No partner
1.0 0.8 0.6 0.4 0.2 0.0
8
4
Uterine: No partner
1.0 0.8 0.6 0.4 0.2 0.0
8
2
Melanoma: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Uterine: Partner
1.0 0.8 0.6 0.4 0.2 0.0
0
10
2.2.2.1. Interaction models. Evidence for the interaction between sex and partner status was assessed by including appropriate second-order terms in the fully adjusted main effect models with sex as a covariate. All models were fitted with stpm2 package [31].
Colorectal: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Melanoma: Partner
1.0 0.8 0.6 0.4 0.2 0.0
0
10
covariates with likelihood ratio tests, but this only supported the inclusion of age as a time-dependent covariate in the final models through additional spline variables. For each site, the optimal number of knots for both the baseline hazard and time dependent effects were determined based on the Bayesian Information Criterion (BIC). Hazard ratios and cumulative probability functions were estimated from models stratified by sex due to convergence issues when including sex as a main effect for individual sites.
All: No partner
1.0 0.8 0.6 0.4 0.2 0.0 8
19
0
2
4
6
8
10
Head and neck: No partner
0
2
4
6
8
10
Time Since Diagnosis (Years) cancer−specific
other cancers
non−cancers
Fig. 2. Stacked plots of cumulative probabilities of death by cancer site and partner status among females, Queensland 1996–2012. These plots show the cumulative probability of death for cancer-specific, other cancers and non-cancers causes stacked on top of each other for each evaluated cancer site.
20
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
2.2.3. Competing risks analysis Competing risk analyses within a flexible parametric framework allow model based smooth estimates of both the impact of covariates on cause-specific hazard rates and an absolute measure of the cause-specific cumulative probabilities of death through cumulative incidence functions. This differs from the more typical Cox models for estimating cause-specific hazards or Fine and Gray competing-risks regression for modelling covariate effects on the cumulative probabilities of death [33,34]. Age was modelled as a categorical variable (20–54, 55–64, 65– 74 and 75+ years) to reduce the number and complexity of calculations, however the estimated cause-specific hazard ratios were similar to the models (not presented) using age as a continuous variable. Cause-specific hazard ratios (HR) for the partner effect were obtained directly from the flexible parametric models. The cumulative probabilities of death (and associated confidence intervals) were estimated through a transformation of these cause-
specific hazards using the post-estimation command stpm2cif [35]. 3. Results Out of 176,050 patients in the final cohort, 68% had a partner. Partnered patients were, on average, three years younger at diagnosis than unpartnered patients (median age 63 versus 66 years), were more likely to be males (61% versus 39%, p < 0.001) and had a lower 10-year cumulative probability of death (46% versus 58% for males and 40% versus 48% for females) over combined sites (Table 1). 3.1. Competing risks analysis Results are presented in each sub-section first for combined sites and then by individual cancers.
Table 2 Cause-specific hazard ratios (95% CI in brackets) for partner effect by sex Queensland, 1996–2012. Site
Sex
Cancer-specific HR (no partner: partner)a,b,c
Colorectal
Lung
Melanoma
Breast
Uterus
Prostate
Kidney
Bladder
Non-Hodgkin lymphoma
Head and neck
Overall (combined sites)
Other cancers p*
HR (no partner: partner)a,b,c
<0.001
Interaction Male Female
1.20 [1.14, 1.27] 1.01 [0.92, 1.11]
Interaction Male Female
1.10 [1.07, 1.15] 1.07 [0.98, 1.13]
Interaction Male Female
1.31 [1.16, 1.48] 1.24 [1.02, 1.44]
Interaction Male Female
n.a. 1.30 [1.26, 1.42]
Interaction Male Female
n.a. 1.40 [1.20, 1.63]
Interaction Male Female
1.32 [1.20, 1.41] n.a.
Interaction Male Female
0.99 [0.88, 1.11] 0.98 [0.88, 1.17]
Interaction Male Female
1.16 [1.02, 1.30] 0.98 [0.83, 1.19]
Interaction Male Female
1.31 [1.17, 1.46] 1.01 [0.94, 1.20]
Interaction Male Female
1.74 [1.56, 1.90] 1.37 [1.16, 1.62]
Interaction Male Female
1.46 [1.42, 1.50] 1.21 [1.17, 1.24]
Non-cancer p*
HR (no partner: partner)a,b,c
<0.001 1.03 [0.94, 1.28] 1.26 [0.92, 1.47]
<0.001
<0.001 1.55 [1.40, 1.84] 1.40 [1.19, 1.62] <0.001
0.001 1.22 [0.92, 1.62] 0.99 [0.64, 1.53]
<0.001
1.33 [1.18, 1.50] 1.27 [1.08, 1.59] <0.001
1.06 [0.92, 1.22] 1.46 [1.21, 1.70] n.a.
<0.001 1.28 [1.06, 1.51] 1.40 [1.37, 1.62]
n.a. n.a. 1.36 [1.24, 1.67]
n.a.
n.a. n.a. 1.54 [1.48, 1.90]
n.a. n.a. 1.17 [0.77, 1.92]
n.a.
n.a. n.a. 2.01 [1.52, 2.25]
n.a. 1.22 [1.10, 1.36] n.a.
0.884
n.a. 1.64 [1.52, 1.77] n.a.
0.018 1.40 [1.01, 1.95] 1.04 [0.63, 1.47]
0.002
0.001 1.64 [1.37, 1.97] 1.06 [0.93, 1.53]
0.087 0.95 [0.75, 1.18] 1.14 [0.81, 1.88]
<0.001
0.003 1.22 [1.04, 1.38] 1.08 [0.98, 1.67] <0.001
0.199 1.09 [0.81, 1.46] 1.01 [0.78, 1.58]
<0.001
1.69 [1.41, 2.02] 1.21 [0.97, 1.69] <0.001
0.007 1.24 [1.01, 1.52] 1.13 [0.75, 1.70]
<0.001
1.96 [1.64, 2.20] 1.84 [1.49, 2.37] <0.001
1.18 [1.11, 1.29] 1.32 [1.24, 1.49]
p*
<0.001 1.44 [1.37, 1.58] 1.41 [1.34, 1.55]
CI Confidence Interval, HR Hazard ratio. a Hazard ratios in normal bold text indicates significant associations existed (95% CI does not include 1). b Models also adjusted for age group at diagnosis. Refer to text for further details. c Reference category is having a partner. * p Values (Wald test, p 0.100 criteria for significance) for interaction term tests whether there is a significant difference in the hazard ratio for partner effect by sex. Refer to text for further details.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
3.1.1. Total cumulative probability of death The estimated 10-year total cumulative probability of death from competing-risks models (Table 1) showed the consistently poorer survival for unpartnered patients with the probability of death over combined sites being 26% higher for unpartnered males and 20% higher for unpartnered females than their partnered counterparts. The survival advantage of being partnered was evident across all individual cancers although the magnitude of the effect varied by specific combination of site and sex, with the survival advantage for partnered versus unpartnered males ranging from 2% (lung) to 30% (head and neck) and for females from 2% (kidney and lung cancer) to 41% (uterus). 3.1.2. Cause-specific cumulative probabilities A comparison of the estimated 10-year cause-specific cumulative probabilities of death also highlighted the survival advantage for partnered cancer patients (Table 1). Stacked plots of the causespecific cumulative probability of death as a function of time (years) since diagnosis give a visual presentation of the total cumulative probability of death broken down by the different causes by site and partner status for males (Fig. 1) and females (Fig. 2). Both unpartnered males and females had a higher probability of cancer-specific death over combined sites than their partnered counterparts (39% versus 30% for males and 33% versus 29% for females) and for non-cancer deaths (14% versus 11% for males and 11% versus 8% for females). However, the corresponding probabilities for other cancer mortality by partner status was similar for males (5% for both subgroups) and females (4% versus 3%). These patterns were also consistent for the majority of individual cancers although the magnitude of the partner effect varied by specific combination of site, sex and cause of death. Exceptions were kidney cancer where the cancer-specific probability of death was either lower or equivalent for unpartnered versus partnered males (31% versus 33%) or females (34% for both subgroups) and for females with either bladder cancer (29% versus 30% respectively) or non-Hodgkin Lymphoma (28% for both subgroups). Both males and females had a substantially greater burden of mortality due to their primary cancer than for other competing mortality causes for the majority of sites irrespective of their partner status. Exceptions were prostate cancer and melanoma (males only), where the probability of cancer-specific and noncancer deaths was similar for both unpartnered (20% versus 17%
21
and 19% versus 17% respectively) and partnered males (17% versus 15% and 14% versus 13%) (Table 1). A consistent pattern of higher probability of death for both unpartnered males and females was evident across all age groups for each site (Appendix). As an illustrative example for colorectal cancer the total 10-year probability of death was 42% for unpartnered males aged 50–64 and 81% for those aged 75+ compared to 35% and 71% for their partnered counterparts respectively. In addition the total probability of death (over combined sites) for unpartnered males was 37%, 47%, 65% and 82% for those aged below 55, 55–64, 65–74 and 75+ respectively with the corresponding probabilities being 26%, 36%, 51% and 68% for their partnered counterparts. The probability of non-cancer deaths increased with advancing age at diagnosis for each individual site (except lung cancer) while deaths from the primary cancer was the most likely cause of death among the younger age groups regardless of sex or partner status. 3.1.3. Cause-specific hazards Cause-specific age-adjusted hazard ratios for the partner effect (Table 2) indicated that being unpartnered was associated with an increased risk of death, and so poorer survival. However, the direction and magnitude of the specific effects varied by site, sex and the cause of death. Results presented in Tables 1 and 2 were estimated with same models, hence cause-specific hazard ratios correspond to the cumulative probabilities of death. 3.1.4. Stage-adjusted analysis The stage-adjusted cause-specific hazard ratios (Table 3) followed a similar pattern to those from unadjusted analysis (Table 2) for all associations across the three cancers. For example, unpartnered male colorectal cancer patients had consistently higher risks of cancer-specific and non-cancer mortality than their partnered counterparts in both un-adjusted and adjusted analyses Further analysis (results not shown) indicated that the partner effect differed for female breast cancer patients by stage at diagnosis (interaction p = 0.008). While unpartnered women with early stage disease had significantly higher mortality risk from all three competing mortality causes, those with advanced disease had a significantly increased risk of cancer-specific and non-cancer mortality only. There was limited evidence (p > 0.05) that the partner effect on competing mortality causes varied by stage for melanoma or colorectal cancer regardless of sex.
Table 3 Cause-specific hazard ratios for partner effect by sex and stage, Queensland, 1996–2012. Site
Sex
Cancer-specific HR (no partner: partner)a,b,c
Colorectal
Melanoma
Breast
Other cancers p*
HR (no partner: partner)
<0.001
Interaction Male Female
1.19 [1.11, 1.27] 1.01 [0.97, 1.08]
Interaction Males Females
1.27 [1.12, 1.41] 1.15 [0.98, 1.37]
Interaction Male Female
n.a. 1.23 [1.14, 1.32]
Non-cancer p
HR (no partner: partner)
<0.001 1.04 [0.83, 1.29] 1.27 [0.96, 1.56]
<0.001
<0.001 1.62 [1.49, 1.87] 1.38 [1.16, 1.61] <0.001
0.016 1.01 [0.85, 1.21] 1.36 [1.04, 1.79]
n.a.
1.44 [1.25, 1.47] 1.53 [1.37, 1.87] n.a.
n.a. 1.42 [1.22, 1.67]
p
n.a. n.a. 1.66 [1.51, 1.91]
CI Confidence Interval, HR Hazard ratio; n.a: not applicable. a Hazard ratios in normal bold text indicates significant associations existed (95% CI does not include 1). b Models also adjusted for age group at diagnosis and stage. Refer to text for further details. c Reference category is having a partner. * p Values (Wald test, p 0.100 criteria for significance) for interaction term tests whether there is a significant difference in the hazard ratio for partner effect by sex. Refer to text for further details.
22
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
3.1.5. Sensitivity analysis Additional analyses (results not shown) indicated that models were not sensitive to the number of knots based on comparisons of the cause-specific cumulative incidence and hazards functions. 4. Discussion We have systematically evaluated the impact of partner status on survival following a cancer diagnosis using state wide population-based cancer registry data. The results of the competing risks analysis illustrated the beneficial effects of being partnered on survival for cancer patients with unpartnered patients being not only at increased risk of deaths from their diagnosed cancer but also from competing causes. A consistent pattern of higher total estimated 10-year cumulative probability of death was also evident for unpartnered patients across all evaluated cancer sites although the cause-specific patterns varied by site and sex. In addition, the partner effect remained even after adjustment for a measure of spread of disease, albeit for only the three sites where this information was available. Studies to date have typically looked at impact of marital or partner status on net survival [4,10,12,15] that is survival in a hypothetical situation where the diagnosed cancer is the only possible cause of death. However, from the perspective of communicating risks, such measures may not be optimal as they ignore competing mortality causes and provide no information about the cumulative probabilities of death for patients. The cumulative probabilities of death can be readily interpreted and communicated in terms of natural frequencies [36]. For example based on information in Table 1 among 100 unpartnered male colorectal cancer patients, within ten years we predict that 41 will have died of their cancer, 3 from another cancer and 15 from noncancer causes whereas the corresponding values for partnered counterparts are lower, being 36, 4 and 11 respectively. A unique feature of the current study is the use of competing risk models in a flexible parametric framework [21] to estimate the cause-specific hazards and cumulative incidence functions which are informative about the impact of partner status on both the mortality rates and the cumulative probability of death for an individual patient [20]. This methodology also allows one model to be fitted for all different competing causes simultaneously thereby allowing for potential shared covariate effects over various causes while also allowing for the possibility that a covariate may not have the same effect for different causes by fitting appropriate interaction terms [31,32]. The reasons for the apparent survival advantage for partnered patients are unclear but are likely to include economic, psychosocial, environmental and structural factors. Having a partner has for example being linked to a healthier lifestyle; greater financial resources and increased practical and/or social support while undergoing treatment [1,3,6]. There is also evidence that social support which would include support from a partner can influence both choice and adherence to treatment modalities in particular adjuvant therapies, as well as time to diagnosis and receipt of treatment [4–9]. Finally the survival advantage of having a partner may reflect the buffering effect of the increased social support on adverse physical and psychosocial effects of cancer and cancerrelated treatments in addition to biological stress-related and immune responses [1,3,37]. The persistence of the partner effect after adjustment for stage and the survival benefit associated with being partnered on competing mortality causes are suggestive evidence that underlying reasons may be related more to treatment, stress and psychological factors than to detection and diagnostic factors, at least for those cancers for which we had stage information available.
4.1. Limitations Limitations of this study include those inherent to using data from the Queensland Cancer Registry such as the lack of stage information for most cancer types. Furthermore, as is the case with population-based registry data, we lacked detailed information on various factors including those related to cancer screening behaviours, cancer-related health behaviours, treatment and psychosocial health. Hence we were unable to control for these potential confounders in our analyses. Another limitation is that the Queensland Cancer Registry only collects partner status at the time of cancer diagnosis; therefore the currently used categories do not account for partner status being a dynamic factor that could have potentially altered post diagnosis. We are therefore unable to account for impact of temporal changes in partner status during the life course of a cancer patient on survival nor assess the accuracy of the collected information on partner status held by the Queensland Cancer Registry. While estimated cause-specific cumulative probabilities of death are advantageous in terms of better understanding the risk factors and realistic implications for patients, this is at the expense of having to rely on cause of death information. In practice the cause of death may be missing or even if available be misclassified or ambiguous due to multiple causes. The Queensland Cancer Registry uses all available information from hospital records, death certificates, autopsy reports and pathology records, to independently assign cause of death information thereby giving increased confidence in the registered cause of death [25]. 5. Conclusions Our analysis suggests that the subgroup of cancer patients who are without a partner when diagnosed with cancer are at increased risk of mortality within ten years from both their primary cancer and competing mortality causes. As such health professionals managing cancer patients should be aware of the increased mortality risk among unpartnered patients and tailor follow-up accordingly. In particular, it is important to better understand the mechanisms by which having a partner is beneficial following a cancer diagnosis, so that this can inform improvements in cancer management for all people with cancer. Contributors PDB conceived the study. PD performed the analysis. PD and PDB drafted the manuscript. All authors contributed to, read and approved the final manuscript. Funding Professor Gavin Turrell is supported by a NHMRC Senior Research Fellowship (ID 1003710). The NHMRC is an external funding agency that provided funds to conduct this research. They had no input into the content or conclusions of this paper. Conflict of interest Conflicts of interest: none. Ethics approval University of Queensland Social and Behavioural Sciences Ethical Review Committee and Queensland Health.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j. canep.2015.12.009. Ten-year cumulative probability of death by site, partner status, age group and sex. References [1] M. Pinquart, P.R. Duberstein, Associations of social networks with cancer mortality: a meta-analysis, Crit. Rev. Oncol. Hematol. 75 (2010) 122–137. [2] H. Kravdal, A. Syse, Changes over time in the effect of marital status on cancer survival, BMC Public Health 11 (2011) 804. [3] M.S. Rendall, M.M. Weden, M.M. Favreault, H. Waldron, The protective effect of marriage for survival: a review and update, Demography 48 (2011) 481–506. [4] A.A. Aizer, M.H. Chen, E.P. McCarthy, M.L. Mendu, S. Koo, T.J. Wilhite, et al., Marital status and survival in patients with cancer, J. Clin. Oncol. 31 (2013) 3869–3876. [5] G. Inverso, B.A. Mahal, A.A. Aizer, R.B. Donoff, N.G. Chau, R.I. Haddad, Marital status and head and neck cancer outcomes, Cancer 121 (2015) 1273–1278. [6] A. Jatoi, P. Novotny, S. Cassivi, M.M. Clark, D. Midthun, C.A. Patten, et al., Does marital status impact survival and quality of life in patients with non-small cell lung cancer? Observations from the mayo clinic lung cancer cohort, Oncologist 12 (2007) 1456–1463. [7] M.L. Jorgensen, J.M. Young, T.A. Dobbins, M.J. Solomon, Does patient age still affect receipt of adjuvant therapy for colorectal cancer in New South Wales, Australia? J. Geriatr. Oncol. 5 (2014) 323–330. [8] E. Pagano, C. Filippini, D. Di Cuonzo, E. Ruffini, R. Zanetti, S. Rosso, et al., Factors affecting pattern of care and survival in a population-based cohort of nonsmall-cell lung cancer incident cases, Cancer Epidemiol. 34 (2010) 483–489. [9] C.A. Reyes Ortiz, J.L. Freeman, Y.F. Kuo, J.S. Goodwin, The influence of marital status on stage at diagnosis and survival of older persons with melanoma, J. Gerontol. A Biol. Sci. Med. Sci. 62 (2007) 892–898. [10] N. Brusselaers, F. Mattsson, A. Johar, A. Wikman, P. Lagergren, J. Lagergren, et al., Marital status and survival after oesophageal cancer surgery: a population-based nationwide cohort study in Sweden, BMJ Open 4 (2014) . [11] T.J. Dilling, K. Bae, R. Paulus, D. Watkins-Bruner, A.S. Garden, A. Forastiere, et al., Impact of gender, partner status, and race on locoregional failure and overall survival in head and neck cancer patients in three radiation therapy oncology group trials, Int. J. Radiat. Oncol. Biol. Phys. 81 (2011) e101–e109. [12] H. Eriksson, J. Lyth, E. Mansson-Brahme, M. Frohm-Nilsson, C. Ingvar, C. Lindholm, et al., Later stage at diagnosis and worse survival in cutaneous malignant melanoma among men living alone: a nationwide populationbased study from Sweden, J. Clin. Oncol. 32 (2014) 1356–1364. [13] H. Mahdi, S. Kumar, A.R. Munkarah, M. Abdalamir, M. Doherty, R. Swensen, Prognostic impact of marital status on survival of women with epithelial ovarian cancer, Psychooncology 22 (2013) 83–88. [14] C. Osborne, G.V. Ostir, X. Du, M.K. Peek, J.S. Goodwin, The influence of marital status on the stage at diagnosis, treatment, and survival of older women with breast cancer, Breast Cancer Res. Treat. 93 (2005) 41–47. [15] L. Wang, S.E. Wilson, D.B. Stewart, C.S. Hollenbeak, Marital status and colon cancer outcomes in US Surveillance, Epidemiology and End Results registries: does marriage affect cancer survival by gender and stage? Cancer Epidemiol. 35 (2011) 417–422. [16] J.L. Nelles, S.A. Joseph, B.R. Konety, The impact of marriage on bladder cancer mortality, Urol. Oncol. 27 (2009) 263–267.
23
[17] K. Saito-Nakaya, N. Nakaya, T. Akechi, M. Inagaki, M. Asai, K. Goto, et al., Marital status and non-small cell lung cancer survival: the lung cancer database project in Japan, Psychooncology 17 (2008) 869–876. [18] J.D. Sammon, M. Morgan, O. Djahangirian, Q.D. Trinh, M. Sun, K.R. Ghani, et al., Marital status: a gender-independent risk factor for poorer survival after radical cystectomy, BJU Int. 110 (2012) 1301–1309. [19] M.J. Rutherford, P.W. Dickman, P.C. Lambert, Comparison of methods for calculating relative survival in population-based studies, Cancer Epidemiol. 36 (2012) 16–21. [20] P.K. Andersen, R.B. Geskus, T. de Witte, H. Putter, Competing risks in epidemiology: possibilities and pitfalls, Int. J. Epidemiol. 41 (2012) 861–870. [21] S.R. Hinchliffe, P.C. Lambert, Flexible parametric modelling of cause-specific hazards to estimate cumulative incidence functions, BMC Med. Res. Methodol. 13 (2013) 13. [22] P. Dasgupta, D.R. Youlden, P.D. Baade, An analysis of competing mortality risks among colorectal cancer survivors in Queensland, 1996-2009, Cancer Causes Control 24 (2013) 897–909. [23] S.D. Fossa, M. Cvancarova, L.L. Chen, A.L. Allan, J. Oldenburg, D.R. Peterson, et al., Adverse prognostic factors for testicular cancer-specific survival: a population-based study of 27,948 patients, J. Clin. Oncol. 29 (2011) 963–970. [24] B.S. Rose, J.H. Jeong, S.K. Nath, S.M. Lu, L.K. Mell, Population-based study of competing mortality in head and neck cancer, J. Clin. Oncol. 29 (2011) 3503– 3509. [25] Queensland Cancer Registry, Cancer in Queensland: Incidence, Mortality, Survival and Prevalence, 1982 to 2012, QCR, Cancer Council Queensland and Queensland Health, Brisbane, 2014. [26] P.D. Baade, P. Royston, P.H. Youl, M.A. Weinstock, A. Geller, J.F. Aitken, Prognostic survival model for people diagnosed with invasive cutaneous melanoma, BMC Cancer 15 (2015) 27. [27] P.D. Baade, G. Turrell, J.F. Aitken, Geographic remoteness, area-level socioeconomic disadvantage and advanced breast cancer: a cross-sectional, multilevel study, J. Epidemiol. Community Health 65 (2011) 1037–1043. [28] P.D. Baade, P. Dasgupta, J.F. Aitken, G. Turrell, Distance to the closest radiotherapy facility and survival after a diagnosis of rectal cancer in Queensland, Med. J. Aust. 195 (2011) 350–354. [29] L.J. Krnjacki, P.D. Baade, B.M. Lynch, J.F. Aitken, Reliability of collecting colorectal cancer stage information from pathology reports and general practitioners in Queensland, Aust. N. Z. J. Public Health 32 (2008) 378–382. [30] Australian Bureau of Statistics. Marriages and Divorces, Australia, 2012 (cat. no. 3310.0). (http://www.abs.gov.au/ausstats/[email protected]/Products/ 3310.02012ChapterMarriages?OpenDocument) ABS, Canberra, 2012 (accessed 03.02.2015). [31] P.C. Lambert, P. Royston, Further development of flexible parametric models for survival analysis, Stata J. 9 (2009) 265–290. [32] P. Royston, P.C. Lambert, Flexible Parametric Survival Analysis Using Stata: Beyond the Cox Model., Stata Press, Texas, 2011. [33] M.A. Cleves, R.G. Gutierrez, M.W. Gould, Y.V. Marchenko, An Introduction to Survival Analysis using Stata, 3rd ed., Stata Press, 2010 Revised ed.. [34] J.P. Fine, R.J. Gray, A proportional hazards model for the subdistribution of a competing risk, J. Am. Stat. Assoc. 94 (1999) 496–509. [35] S.R. Hinchliffe, P.C. Lambert, Extending the flexible parametric survival model for competing risks, Stata J. 13 (2013) 344–355. [36] S. Eloranta, J. Adolfsson, P.C. Lambert, P. Stattin, O. Akre, T.M.L. Andersson, et al., How can we make cancer survival statistics more useful for patients and clinicians: an illustration using localized prostate cancer in Sweden, Cancer Causes Control 24 (2013) 505–515. [37] J. Giese-Davis, A. Waller, L.E. Carlson, S. Groff, L. Zhong, E. Neri, et al., Screening for distress, the 6th vital sign: common problems in cancer outpatients over one year in usual care: associations with marital status, sex, and age, BMC Cancer 12 (2012) 441.
Contents lists available at ScienceDirect
Cancer Epidemiology The International Journal of Cancer Epidemiology, Detection, and Prevention journal homepage: www.cancerepidemiology.net
Partner status and survival after cancer: A competing risks analysis Paramita Dasguptaa , Gavin Turrellb , Joanne F. Aitkena,b,c , Peter D. Baadea,b,d,* a
Cancer Council Queensland, P.O. Box 201, Spring Hill, QLD 4004, Australia School of Public Health and Social Work, Queensland University of Technology, Herston Road, Kelvin Grove, QLD 4059, Australia c School of Population Health, University of Queensland, Brisbane, Australia d Menzies Health Institute Queensland, Griffith University, Gold Coast Campus, Parklands Drive, Southport, QLD 4222, Australia b
A R T I C L E I N F O
A B S T R A C T
Article history: Received 22 October 2015 Received in revised form 13 December 2015 Accepted 18 December 2015 Available online 14 January 2016
Objective: The survival benefits of having a partner for all cancers combined is well recognized, however its prognostic importance for individual cancer types, including competing mortality causes, is less clear. This study was undertaken to quantify the impact of partner status on survival due to cancer-specific and competing mortality causes. Methods: Data were obtained from the population-based Queensland Cancer Registry on 176,050 incident cases of ten leading cancers diagnosed in Queensland (Australia) from 1996 to 2012. Flexible parametric competing-risks models were used to estimate cause-specific hazards and cumulative probabilities of death, adjusting for age, stage (breast, colorectal and melanoma only) and stratifying by sex. Results: Both unpartnered males and females had higher total cumulative probability of death than their partnered counterparts for each site. For example, the survival disadvantage for unpartnered males ranged from 3% to 30% with higher mortality burden from both the primary cancer and competing mortality causes. The cause-specific age-adjusted hazard ratios were also consistent with patients without a partner having increased mortality risk although the specific effect varied by site, sex and cause of death. For all combined sites, unpartnered males had a 46%, 18% and 44% higher risk of cancer-specific, other cancer and non-cancer mortality respectively with similar patterns for females. The higher mortality risk persisted after adjustment for stage. Conclusions: It is important to better understand the mechanisms by which having a partner is beneficial following a cancer diagnosis, so that this can inform improvements in cancer management for all people with cancer. ã 2015 Elsevier Ltd. All rights reserved.
Keywords: Cancer Survival Partner status Inequalities Competing risks
1. Introduction The protective effect of marriage on overall cancer survival [1] has been well established. Emerging evidence also suggests that the survival benefit for married patients may be increasing over time [2]. The protective effect of marriage may reflect increased social support over the cancer continuum from diagnosis to survivorship to buffer the adverse effects of stress-related biological processes and immune responses [1,3]. Having a partner may also provide a source of financial, practical and emotional support while undergoing treatment and has been linked to more timely care, increased receipt of curative therapies, greater
* Corresponding author at: Cancer Council Queensland, P.O. Box 201, Spring Hill, QLD 4004, Australia. Fax: +61 7 3259 8527. E-mail addresses: [email protected] (P. Dasgupta), [email protected] (G. Turrell), [email protected] (J.F. Aitken), [email protected] (P.D. Baade). http://dx.doi.org/10.1016/j.canep.2015.12.009 1877-7821/ ã 2015 Elsevier Ltd. All rights reserved.
compliance with multi-modal treatments and improved psychosocial outcomes [1,3–9]. Results from studies that combine cancer sites may be affected by too much aggregation across different diseases. However findings to date on the impact of being married on cause-specific survival for individual cancer sites are equivocal with nonsignificant [6,10], protective [4,5,11–15] and mixed effects [16,17] being reported after adjusting for various combinations of known survival determinants. Some [4,12,16] but not all [15,18] studies also reported that the partner effect varied by gender. These inconsistencies likely reflect the heterogeneity in time period, cohort characteristic, covariates included in statistical models and definition of marital status across studies. While several categorized patients as married or unmarried based on official marital status categories [4,5,13,14] or extended the married group to also include partnered patients [11,12], others split the unmarried category into sub-groups (single, widowed, divorced or separated) [6,15–17].
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
The impact of partner status has typically being reported using net survival [4,10,12,15], describing the hypothetical situation where the diagnosed cancer is the only cause of death [19]. In the real world, however patients are at risk of death from various mutually exclusive causes. Hence partitioning overall mortality into different causes has the potential to provide a richer understanding of the impact having a partner has on survival after a cancer diagnosis [20,21]. To date, only a few studies have reported on the beneficial effects of having a partner on both cancer-specific [5,22–24] and competing mortality causes [22,24], however none have systematically compared these effects across multiple cancer sites. In this study we quantified the survival benefit of being partnered for each of the ten leading cancers diagnosed in Queensland, Australia using population-based data and competing risk methods [21]. 2. Methods Approval for the use of these de-identified data was obtained from the data custodian, Queensland Health. In addition, ethics and
17
data custodian approval for the extraction of stage information was obtained by The University of Queensland Social and Behavioural Sciences Ethical Review Committee and Queensland Health. Data were extracted from the population-based Queensland Cancer Registry, to which notifications of any cancer diagnosis (except keratinocyte cancers) is a statutory requirement [25]. All incident cases of the ten leading cancers diagnosed in Queensland from 1 January 1996 to 31st December 2012 (Table 1) were initially considered. Cases identified at death or autopsy, diagnosed with multiple primary cancers, with unknown marital status or who survived for less than one day were excluded. The cohort was restricted to patients aged 20–79 years at diagnosis consistent with an earlier published study [22]. While the Queensland Cancer Registry does not habitually gather information pertaining to stage at diagnosis, additional clinical information collected routinely [25] or manually extracted from pathology forms enabled a measure of stage at diagnosis to be calculated for melanoma [26], breast [27] and colorectal cancer [28,29]. To allow sufficient numbers for modelling, cases were collectively grouped as early (localized/locally advanced) or
Table 1 Demographic characteristics by partner status, sex and mortality outcomes for selected cancers, Queensland, 1996–2012. Site (Partner)
Males N
Females 10-year cumulative probability of death (%)
N
Totala
Cancer-specific
Other cancers
Non-cancer
10-year cumulative probability of death (%) Totala
Cancer-specific
Other cancers
Non-cancer
Colorectal No partner Has partner
4,654 12,930
59 [54,65] 51 [46,55]
41 [39,44] 36 [34,38]
3 [2,4] 4 [3,5]
15 [13,17] 11 [9,12]
5,649 7,413
53 [50,58] 46 [41,51]
40 [39,42] 36 [34,39]
3 [2,4] 2 [1,3]
10 [9,12] 8 [6,9]
Lung No partner Has partner
4,797 9,094
92 [88,97] 90 [86,94]
85 [83,87] 83 [81,85]
1 [0,2] 1 [0,2]
6 [5,8] 6 [5,7]
3,576 4,058
87 [82,93] 85 [80,91]
81 [78,84] 80 [77,83]
1 [0,2] 1 [0,2]
6 [4,7] 4 [3,6]
Melanoma No partner Has partner
3,221 9,454
44 [36,48] 35 [32,40]
19 [16,21] 14 [13,16]
8 [6,9] 8 [7,9]
17 [14,18] 13 [12,15]
3,222 5,922
30 [24,36] 22 [17,28]
11 [9,13] 9 [7,11]
6 [4,8] 4 [3,6]
13 [11,15] 9 [7,11]
Breast-female No partner Has partner
11,345 22,872
37 [33,41] 28 [24,31]
20 [19,22] 17 [15,18]
4 [3,5] 3 [2,4]
13 [11,14] 8 [7,9]
Uterus No partner Has partner
1,816 2,738
35 [26,43] 25 [18,31]
20 [16,23] 15 [12,17]
3 [1,5] 3 [1,5]
12 [9,15] 7 [5,9]
Prostate No partner Has partner
7,728 27,984
43 [38,49] 37 [33,41]
20 [18,23] 17 [15,19]
6 [5,7] 5 [4,6]
17 [15,19] 15 [14,16]
Kidney No partner Has partner
1,085 3,120
55 [45,66] 49 [40,57]
31 [27,36] 33 [29,37]
7 [4,9] 5 [3,7]
17 [14,21] 11 [8,13]
988 1,340
49 [38,63] 48 [40,55]
34 [27,40] 34 [30,36]
3 [2,7] 3 [2,5]
12 [9,16] 11 [8,14]
Bladder No partner Has partner
1,183 2,789
52 [42,63] 48 [41,57]
27 [23,32] 24 [21,28]
9 [6,11] 10 [8,12]
16 [13,20] 14 [12,17]
608 580
47 [31,62] 45 [36,59]
29 [21,36] 30 [26,38]
6 [2,9] 5 [3,7]
12 [8,17] 10 [7,14]
Non-Hodgkin Lymphoma No partner Has partner
1,182 3,499
55 [45,66] 45 [38,52]
36 [32,41] 29 [26,32]
5 [3,8] 6 [4,8]
14 [10,17] 10 [8,12]
1,447 2,202
42 [33,51] 40 [31,48]
28 [24,32] 28 [23,31]
4 [2,6] 4 [2,6]
10 [7,13] 8 [6,11]
Head and neck No partner Has partner
2,363 3,445
74 [65,83] 57 [48,65]
49 [45,53] 35 [31,38]
7 [5,9] 9 [6,11]
18 [15,21] 13 [11,16]
783 963
63 [46,78] 50 [36,65]
42 [34,49] 34 [28,41]
7 [3,10] 7 [3,11]
14 [9,19] 9 [5,13]
Overall (combined sites) No partner Has partner
26,213 72,315
58 [55,62] 46 [42,49]
39 [38,40] 30 [29,31]
5 [4,7] 5 [3,6]
14 [13,15] 11 [10,12]
29,434 48,088
48 [45,51] 40 [37,43]
33 [32,34] 29 [28,30]
4 [3,5] 3 [2,4]
11 [10,12] 8 [7,9]
a Sum of the three cause-specific probabilities of death. Cancer-specific are deaths due to primary cancer, other cancers are deaths due to any other cancers. Refer to text for further details.
18
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
advanced (regional, distant) for melanoma and colorectal cancer. For breast cancer, cases were defined as ‘early’ (Stage I) if 20 mm size with no evident nodal spread or metastases while Stage II, III and IV cancers which could not be distinguished based on available information were collectively categorized as ‘advanced’ [27]. For all three sites, cases with unknown stage were excluded. Information about marital status is collected routinely by the Queensland Cancer Registry, based on a patients self-reported data as noted in the medical record. For this study cases with a partner included those who were married or in a de facto relationship.
2.2. Statistical analysis All statistical analyses were performed with Stata/SE version 13 (StataCorp, TX, USA) and were stratified by sex due to differences in the age-specific patterns of partner status by sex [30].
2.1. Survival times The cohort was followed up to 31st December 2012 with matching to the Queensland Registrar of Births, Deaths and Marriages and the National Death Index [25] to identify
All: Partner
1.0 0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
2.2.1. Exploratory analysis Cohort characteristics were compared using the x2 test.
All: No partner
1.0
0.0
0.0 0
2
4
6
8
10
Colorectal: Partner
1.0
Probability of Death
Queensland-specific and interstate deaths respectively. Survival was measured in days from the date of diagnosis to death or the study end point, with cases alive at the study end point being censored. Cause of death information from the Queensland Cancer Registry [25] was categorized as cancer-specific if attributed to the primary cancer, other cancers or non-cancers. Follow-up was restricted to 10 years after diagnosis.
0
2
4
6
8
10
Colorectal: No partner
1.0
Lung: Partner
1.0
0.8
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
0
2
4
6
8
10
Melanoma: Partner
1.0
0
2
4
6
8
10
Melanoma: No partner
1.0
0.0 0
2
4
6
8
10
Prostate: Partner
1.0
0
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
2
4
6
8
10
Kidney: Partner
1.0
0
2
4
6
8
10
Kidney: No partner
1.0
2
4
6
8
10
Bladder: Partner
1.0
0
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
4
6
8
10
NHL: Partner
1.0
0
2
4
6
8
10
NHL: No partner
1.0
2
4
6
8
10
Head and neck: Partner
1.0
1.0
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0.0
0.0
0.0
2
4
6
8
10
0
2
4
6
8
10
8
10
2
4
6
8
10
0.0 0
0.8
0
6
Bladder: No partner
1.0
0.6
2
4
0.0 0
0.8
0
2
Prostate: No partner
1.0
0.8
0
Lung: No partner
1.0
0
2
4
6
8
10
Head and neck: No partner
0.0 0
2
4
6
8
10
0
2
4
6
8
10
Time Since Diagnosis (Years) cancer−specific
other cancers
non−cancers
Fig 1. Stacked plots of cumulative probabilities of death by cancer site and partner status among males, Queensland 1996–2012. These plots show the cumulative probability of death due to cancer-specific, other cancers and non-cancers causes stacked on top of each other for each evaluated cancer site.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
2.2.2. Flexible parametric models Survival analyses were carried out within a flexible parametric modelling framework which use restricted cubic splines to model the baseline hazard function [31,32]. For each site, flexible parametric models on the log cumulative hazard scale were used to model the three competing mortality causes simultaneously [21,32]. Additional subgroup analysis by cancer stage were carried out for melanoma, colorectal and breast cancer. Models were adjusted for age, partner status and cancer stage where relevant. Baseline hazard functions were estimated separately for each cause by using the rcsbaseoff option during model fitting and including the three cause-specific indicators as main effects and time-dependent effects [32]. Interaction terms between each cause of death and age and partner status and stage were fitted. Time-dependent effects were assessed for all
All: Partner
1.0 0.8 0.6 0.4 0.2 0.0
Probability of Death
0
2
4
6
10
Colorectal: Partner
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
8
0
2
4
6
8
0
2
4
6
10
Bladder: Partner
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
0
0
10
Head and neck: Partner
1.0 0.8 0.6 0.4 0.2 0.0
1.0 0.8 0.6 0.4 0.2 0.0 0
2
4
6
8
10
6
8
10
2
4
6
8
2
4
6
8
2
4
6
8
0
2
4
6
8
Lung: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
0
10
0
2
0
2
6
8
10
4
6
8
4
6
8
0
2
4
6
0
10
0
0
10
6
8
10
2
4
6
8
10
2
4
6
8
10
NHL: No partner
1.0 0.8 0.6 0.4 0.2 0.0 8
4
Kidney: No partner
1.0 0.8 0.6 0.4 0.2 0.0 10
2
Breast: No partner
1.0 0.8 0.6 0.4 0.2 0.0
NHL: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
4
Kidney: Partner
1.0 0.8 0.6 0.4 0.2 0.0 10
2
Lung: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Breast: Partner
1.0 0.8 0.6 0.4 0.2 0.0
Bladder: No partner
1.0 0.8 0.6 0.4 0.2 0.0
8
4
Uterine: No partner
1.0 0.8 0.6 0.4 0.2 0.0
8
2
Melanoma: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Uterine: Partner
1.0 0.8 0.6 0.4 0.2 0.0
0
10
2.2.2.1. Interaction models. Evidence for the interaction between sex and partner status was assessed by including appropriate second-order terms in the fully adjusted main effect models with sex as a covariate. All models were fitted with stpm2 package [31].
Colorectal: No partner
1.0 0.8 0.6 0.4 0.2 0.0
Melanoma: Partner
1.0 0.8 0.6 0.4 0.2 0.0
0
10
covariates with likelihood ratio tests, but this only supported the inclusion of age as a time-dependent covariate in the final models through additional spline variables. For each site, the optimal number of knots for both the baseline hazard and time dependent effects were determined based on the Bayesian Information Criterion (BIC). Hazard ratios and cumulative probability functions were estimated from models stratified by sex due to convergence issues when including sex as a main effect for individual sites.
All: No partner
1.0 0.8 0.6 0.4 0.2 0.0 8
19
0
2
4
6
8
10
Head and neck: No partner
0
2
4
6
8
10
Time Since Diagnosis (Years) cancer−specific
other cancers
non−cancers
Fig. 2. Stacked plots of cumulative probabilities of death by cancer site and partner status among females, Queensland 1996–2012. These plots show the cumulative probability of death for cancer-specific, other cancers and non-cancers causes stacked on top of each other for each evaluated cancer site.
20
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
2.2.3. Competing risks analysis Competing risk analyses within a flexible parametric framework allow model based smooth estimates of both the impact of covariates on cause-specific hazard rates and an absolute measure of the cause-specific cumulative probabilities of death through cumulative incidence functions. This differs from the more typical Cox models for estimating cause-specific hazards or Fine and Gray competing-risks regression for modelling covariate effects on the cumulative probabilities of death [33,34]. Age was modelled as a categorical variable (20–54, 55–64, 65– 74 and 75+ years) to reduce the number and complexity of calculations, however the estimated cause-specific hazard ratios were similar to the models (not presented) using age as a continuous variable. Cause-specific hazard ratios (HR) for the partner effect were obtained directly from the flexible parametric models. The cumulative probabilities of death (and associated confidence intervals) were estimated through a transformation of these cause-
specific hazards using the post-estimation command stpm2cif [35]. 3. Results Out of 176,050 patients in the final cohort, 68% had a partner. Partnered patients were, on average, three years younger at diagnosis than unpartnered patients (median age 63 versus 66 years), were more likely to be males (61% versus 39%, p < 0.001) and had a lower 10-year cumulative probability of death (46% versus 58% for males and 40% versus 48% for females) over combined sites (Table 1). 3.1. Competing risks analysis Results are presented in each sub-section first for combined sites and then by individual cancers.
Table 2 Cause-specific hazard ratios (95% CI in brackets) for partner effect by sex Queensland, 1996–2012. Site
Sex
Cancer-specific HR (no partner: partner)a,b,c
Colorectal
Lung
Melanoma
Breast
Uterus
Prostate
Kidney
Bladder
Non-Hodgkin lymphoma
Head and neck
Overall (combined sites)
Other cancers p*
HR (no partner: partner)a,b,c
<0.001
Interaction Male Female
1.20 [1.14, 1.27] 1.01 [0.92, 1.11]
Interaction Male Female
1.10 [1.07, 1.15] 1.07 [0.98, 1.13]
Interaction Male Female
1.31 [1.16, 1.48] 1.24 [1.02, 1.44]
Interaction Male Female
n.a. 1.30 [1.26, 1.42]
Interaction Male Female
n.a. 1.40 [1.20, 1.63]
Interaction Male Female
1.32 [1.20, 1.41] n.a.
Interaction Male Female
0.99 [0.88, 1.11] 0.98 [0.88, 1.17]
Interaction Male Female
1.16 [1.02, 1.30] 0.98 [0.83, 1.19]
Interaction Male Female
1.31 [1.17, 1.46] 1.01 [0.94, 1.20]
Interaction Male Female
1.74 [1.56, 1.90] 1.37 [1.16, 1.62]
Interaction Male Female
1.46 [1.42, 1.50] 1.21 [1.17, 1.24]
Non-cancer p*
HR (no partner: partner)a,b,c
<0.001 1.03 [0.94, 1.28] 1.26 [0.92, 1.47]
<0.001
<0.001 1.55 [1.40, 1.84] 1.40 [1.19, 1.62] <0.001
0.001 1.22 [0.92, 1.62] 0.99 [0.64, 1.53]
<0.001
1.33 [1.18, 1.50] 1.27 [1.08, 1.59] <0.001
1.06 [0.92, 1.22] 1.46 [1.21, 1.70] n.a.
<0.001 1.28 [1.06, 1.51] 1.40 [1.37, 1.62]
n.a. n.a. 1.36 [1.24, 1.67]
n.a.
n.a. n.a. 1.54 [1.48, 1.90]
n.a. n.a. 1.17 [0.77, 1.92]
n.a.
n.a. n.a. 2.01 [1.52, 2.25]
n.a. 1.22 [1.10, 1.36] n.a.
0.884
n.a. 1.64 [1.52, 1.77] n.a.
0.018 1.40 [1.01, 1.95] 1.04 [0.63, 1.47]
0.002
0.001 1.64 [1.37, 1.97] 1.06 [0.93, 1.53]
0.087 0.95 [0.75, 1.18] 1.14 [0.81, 1.88]
<0.001
0.003 1.22 [1.04, 1.38] 1.08 [0.98, 1.67] <0.001
0.199 1.09 [0.81, 1.46] 1.01 [0.78, 1.58]
<0.001
1.69 [1.41, 2.02] 1.21 [0.97, 1.69] <0.001
0.007 1.24 [1.01, 1.52] 1.13 [0.75, 1.70]
<0.001
1.96 [1.64, 2.20] 1.84 [1.49, 2.37] <0.001
1.18 [1.11, 1.29] 1.32 [1.24, 1.49]
p*
<0.001 1.44 [1.37, 1.58] 1.41 [1.34, 1.55]
CI Confidence Interval, HR Hazard ratio. a Hazard ratios in normal bold text indicates significant associations existed (95% CI does not include 1). b Models also adjusted for age group at diagnosis. Refer to text for further details. c Reference category is having a partner. * p Values (Wald test, p 0.100 criteria for significance) for interaction term tests whether there is a significant difference in the hazard ratio for partner effect by sex. Refer to text for further details.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
3.1.1. Total cumulative probability of death The estimated 10-year total cumulative probability of death from competing-risks models (Table 1) showed the consistently poorer survival for unpartnered patients with the probability of death over combined sites being 26% higher for unpartnered males and 20% higher for unpartnered females than their partnered counterparts. The survival advantage of being partnered was evident across all individual cancers although the magnitude of the effect varied by specific combination of site and sex, with the survival advantage for partnered versus unpartnered males ranging from 2% (lung) to 30% (head and neck) and for females from 2% (kidney and lung cancer) to 41% (uterus). 3.1.2. Cause-specific cumulative probabilities A comparison of the estimated 10-year cause-specific cumulative probabilities of death also highlighted the survival advantage for partnered cancer patients (Table 1). Stacked plots of the causespecific cumulative probability of death as a function of time (years) since diagnosis give a visual presentation of the total cumulative probability of death broken down by the different causes by site and partner status for males (Fig. 1) and females (Fig. 2). Both unpartnered males and females had a higher probability of cancer-specific death over combined sites than their partnered counterparts (39% versus 30% for males and 33% versus 29% for females) and for non-cancer deaths (14% versus 11% for males and 11% versus 8% for females). However, the corresponding probabilities for other cancer mortality by partner status was similar for males (5% for both subgroups) and females (4% versus 3%). These patterns were also consistent for the majority of individual cancers although the magnitude of the partner effect varied by specific combination of site, sex and cause of death. Exceptions were kidney cancer where the cancer-specific probability of death was either lower or equivalent for unpartnered versus partnered males (31% versus 33%) or females (34% for both subgroups) and for females with either bladder cancer (29% versus 30% respectively) or non-Hodgkin Lymphoma (28% for both subgroups). Both males and females had a substantially greater burden of mortality due to their primary cancer than for other competing mortality causes for the majority of sites irrespective of their partner status. Exceptions were prostate cancer and melanoma (males only), where the probability of cancer-specific and noncancer deaths was similar for both unpartnered (20% versus 17%
21
and 19% versus 17% respectively) and partnered males (17% versus 15% and 14% versus 13%) (Table 1). A consistent pattern of higher probability of death for both unpartnered males and females was evident across all age groups for each site (Appendix). As an illustrative example for colorectal cancer the total 10-year probability of death was 42% for unpartnered males aged 50–64 and 81% for those aged 75+ compared to 35% and 71% for their partnered counterparts respectively. In addition the total probability of death (over combined sites) for unpartnered males was 37%, 47%, 65% and 82% for those aged below 55, 55–64, 65–74 and 75+ respectively with the corresponding probabilities being 26%, 36%, 51% and 68% for their partnered counterparts. The probability of non-cancer deaths increased with advancing age at diagnosis for each individual site (except lung cancer) while deaths from the primary cancer was the most likely cause of death among the younger age groups regardless of sex or partner status. 3.1.3. Cause-specific hazards Cause-specific age-adjusted hazard ratios for the partner effect (Table 2) indicated that being unpartnered was associated with an increased risk of death, and so poorer survival. However, the direction and magnitude of the specific effects varied by site, sex and the cause of death. Results presented in Tables 1 and 2 were estimated with same models, hence cause-specific hazard ratios correspond to the cumulative probabilities of death. 3.1.4. Stage-adjusted analysis The stage-adjusted cause-specific hazard ratios (Table 3) followed a similar pattern to those from unadjusted analysis (Table 2) for all associations across the three cancers. For example, unpartnered male colorectal cancer patients had consistently higher risks of cancer-specific and non-cancer mortality than their partnered counterparts in both un-adjusted and adjusted analyses Further analysis (results not shown) indicated that the partner effect differed for female breast cancer patients by stage at diagnosis (interaction p = 0.008). While unpartnered women with early stage disease had significantly higher mortality risk from all three competing mortality causes, those with advanced disease had a significantly increased risk of cancer-specific and non-cancer mortality only. There was limited evidence (p > 0.05) that the partner effect on competing mortality causes varied by stage for melanoma or colorectal cancer regardless of sex.
Table 3 Cause-specific hazard ratios for partner effect by sex and stage, Queensland, 1996–2012. Site
Sex
Cancer-specific HR (no partner: partner)a,b,c
Colorectal
Melanoma
Breast
Other cancers p*
HR (no partner: partner)
<0.001
Interaction Male Female
1.19 [1.11, 1.27] 1.01 [0.97, 1.08]
Interaction Males Females
1.27 [1.12, 1.41] 1.15 [0.98, 1.37]
Interaction Male Female
n.a. 1.23 [1.14, 1.32]
Non-cancer p
HR (no partner: partner)
<0.001 1.04 [0.83, 1.29] 1.27 [0.96, 1.56]
<0.001
<0.001 1.62 [1.49, 1.87] 1.38 [1.16, 1.61] <0.001
0.016 1.01 [0.85, 1.21] 1.36 [1.04, 1.79]
n.a.
1.44 [1.25, 1.47] 1.53 [1.37, 1.87] n.a.
n.a. 1.42 [1.22, 1.67]
p
n.a. n.a. 1.66 [1.51, 1.91]
CI Confidence Interval, HR Hazard ratio; n.a: not applicable. a Hazard ratios in normal bold text indicates significant associations existed (95% CI does not include 1). b Models also adjusted for age group at diagnosis and stage. Refer to text for further details. c Reference category is having a partner. * p Values (Wald test, p 0.100 criteria for significance) for interaction term tests whether there is a significant difference in the hazard ratio for partner effect by sex. Refer to text for further details.
22
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
3.1.5. Sensitivity analysis Additional analyses (results not shown) indicated that models were not sensitive to the number of knots based on comparisons of the cause-specific cumulative incidence and hazards functions. 4. Discussion We have systematically evaluated the impact of partner status on survival following a cancer diagnosis using state wide population-based cancer registry data. The results of the competing risks analysis illustrated the beneficial effects of being partnered on survival for cancer patients with unpartnered patients being not only at increased risk of deaths from their diagnosed cancer but also from competing causes. A consistent pattern of higher total estimated 10-year cumulative probability of death was also evident for unpartnered patients across all evaluated cancer sites although the cause-specific patterns varied by site and sex. In addition, the partner effect remained even after adjustment for a measure of spread of disease, albeit for only the three sites where this information was available. Studies to date have typically looked at impact of marital or partner status on net survival [4,10,12,15] that is survival in a hypothetical situation where the diagnosed cancer is the only possible cause of death. However, from the perspective of communicating risks, such measures may not be optimal as they ignore competing mortality causes and provide no information about the cumulative probabilities of death for patients. The cumulative probabilities of death can be readily interpreted and communicated in terms of natural frequencies [36]. For example based on information in Table 1 among 100 unpartnered male colorectal cancer patients, within ten years we predict that 41 will have died of their cancer, 3 from another cancer and 15 from noncancer causes whereas the corresponding values for partnered counterparts are lower, being 36, 4 and 11 respectively. A unique feature of the current study is the use of competing risk models in a flexible parametric framework [21] to estimate the cause-specific hazards and cumulative incidence functions which are informative about the impact of partner status on both the mortality rates and the cumulative probability of death for an individual patient [20]. This methodology also allows one model to be fitted for all different competing causes simultaneously thereby allowing for potential shared covariate effects over various causes while also allowing for the possibility that a covariate may not have the same effect for different causes by fitting appropriate interaction terms [31,32]. The reasons for the apparent survival advantage for partnered patients are unclear but are likely to include economic, psychosocial, environmental and structural factors. Having a partner has for example being linked to a healthier lifestyle; greater financial resources and increased practical and/or social support while undergoing treatment [1,3,6]. There is also evidence that social support which would include support from a partner can influence both choice and adherence to treatment modalities in particular adjuvant therapies, as well as time to diagnosis and receipt of treatment [4–9]. Finally the survival advantage of having a partner may reflect the buffering effect of the increased social support on adverse physical and psychosocial effects of cancer and cancerrelated treatments in addition to biological stress-related and immune responses [1,3,37]. The persistence of the partner effect after adjustment for stage and the survival benefit associated with being partnered on competing mortality causes are suggestive evidence that underlying reasons may be related more to treatment, stress and psychological factors than to detection and diagnostic factors, at least for those cancers for which we had stage information available.
4.1. Limitations Limitations of this study include those inherent to using data from the Queensland Cancer Registry such as the lack of stage information for most cancer types. Furthermore, as is the case with population-based registry data, we lacked detailed information on various factors including those related to cancer screening behaviours, cancer-related health behaviours, treatment and psychosocial health. Hence we were unable to control for these potential confounders in our analyses. Another limitation is that the Queensland Cancer Registry only collects partner status at the time of cancer diagnosis; therefore the currently used categories do not account for partner status being a dynamic factor that could have potentially altered post diagnosis. We are therefore unable to account for impact of temporal changes in partner status during the life course of a cancer patient on survival nor assess the accuracy of the collected information on partner status held by the Queensland Cancer Registry. While estimated cause-specific cumulative probabilities of death are advantageous in terms of better understanding the risk factors and realistic implications for patients, this is at the expense of having to rely on cause of death information. In practice the cause of death may be missing or even if available be misclassified or ambiguous due to multiple causes. The Queensland Cancer Registry uses all available information from hospital records, death certificates, autopsy reports and pathology records, to independently assign cause of death information thereby giving increased confidence in the registered cause of death [25]. 5. Conclusions Our analysis suggests that the subgroup of cancer patients who are without a partner when diagnosed with cancer are at increased risk of mortality within ten years from both their primary cancer and competing mortality causes. As such health professionals managing cancer patients should be aware of the increased mortality risk among unpartnered patients and tailor follow-up accordingly. In particular, it is important to better understand the mechanisms by which having a partner is beneficial following a cancer diagnosis, so that this can inform improvements in cancer management for all people with cancer. Contributors PDB conceived the study. PD performed the analysis. PD and PDB drafted the manuscript. All authors contributed to, read and approved the final manuscript. Funding Professor Gavin Turrell is supported by a NHMRC Senior Research Fellowship (ID 1003710). The NHMRC is an external funding agency that provided funds to conduct this research. They had no input into the content or conclusions of this paper. Conflict of interest Conflicts of interest: none. Ethics approval University of Queensland Social and Behavioural Sciences Ethical Review Committee and Queensland Health.
P. Dasgupta et al. / Cancer Epidemiology 41 (2016) 16–23
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j. canep.2015.12.009. Ten-year cumulative probability of death by site, partner status, age group and sex. References [1] M. Pinquart, P.R. Duberstein, Associations of social networks with cancer mortality: a meta-analysis, Crit. Rev. Oncol. Hematol. 75 (2010) 122–137. [2] H. Kravdal, A. Syse, Changes over time in the effect of marital status on cancer survival, BMC Public Health 11 (2011) 804. [3] M.S. Rendall, M.M. Weden, M.M. Favreault, H. Waldron, The protective effect of marriage for survival: a review and update, Demography 48 (2011) 481–506. [4] A.A. Aizer, M.H. Chen, E.P. McCarthy, M.L. Mendu, S. Koo, T.J. Wilhite, et al., Marital status and survival in patients with cancer, J. Clin. Oncol. 31 (2013) 3869–3876. [5] G. Inverso, B.A. Mahal, A.A. Aizer, R.B. Donoff, N.G. Chau, R.I. Haddad, Marital status and head and neck cancer outcomes, Cancer 121 (2015) 1273–1278. [6] A. Jatoi, P. Novotny, S. Cassivi, M.M. Clark, D. Midthun, C.A. Patten, et al., Does marital status impact survival and quality of life in patients with non-small cell lung cancer? Observations from the mayo clinic lung cancer cohort, Oncologist 12 (2007) 1456–1463. [7] M.L. Jorgensen, J.M. Young, T.A. Dobbins, M.J. Solomon, Does patient age still affect receipt of adjuvant therapy for colorectal cancer in New South Wales, Australia? J. Geriatr. Oncol. 5 (2014) 323–330. [8] E. Pagano, C. Filippini, D. Di Cuonzo, E. Ruffini, R. Zanetti, S. Rosso, et al., Factors affecting pattern of care and survival in a population-based cohort of nonsmall-cell lung cancer incident cases, Cancer Epidemiol. 34 (2010) 483–489. [9] C.A. Reyes Ortiz, J.L. Freeman, Y.F. Kuo, J.S. Goodwin, The influence of marital status on stage at diagnosis and survival of older persons with melanoma, J. Gerontol. A Biol. Sci. Med. Sci. 62 (2007) 892–898. [10] N. Brusselaers, F. Mattsson, A. Johar, A. Wikman, P. Lagergren, J. Lagergren, et al., Marital status and survival after oesophageal cancer surgery: a population-based nationwide cohort study in Sweden, BMJ Open 4 (2014) . [11] T.J. Dilling, K. Bae, R. Paulus, D. Watkins-Bruner, A.S. Garden, A. Forastiere, et al., Impact of gender, partner status, and race on locoregional failure and overall survival in head and neck cancer patients in three radiation therapy oncology group trials, Int. J. Radiat. Oncol. Biol. Phys. 81 (2011) e101–e109. [12] H. Eriksson, J. Lyth, E. Mansson-Brahme, M. Frohm-Nilsson, C. Ingvar, C. Lindholm, et al., Later stage at diagnosis and worse survival in cutaneous malignant melanoma among men living alone: a nationwide populationbased study from Sweden, J. Clin. Oncol. 32 (2014) 1356–1364. [13] H. Mahdi, S. Kumar, A.R. Munkarah, M. Abdalamir, M. Doherty, R. Swensen, Prognostic impact of marital status on survival of women with epithelial ovarian cancer, Psychooncology 22 (2013) 83–88. [14] C. Osborne, G.V. Ostir, X. Du, M.K. Peek, J.S. Goodwin, The influence of marital status on the stage at diagnosis, treatment, and survival of older women with breast cancer, Breast Cancer Res. Treat. 93 (2005) 41–47. [15] L. Wang, S.E. Wilson, D.B. Stewart, C.S. Hollenbeak, Marital status and colon cancer outcomes in US Surveillance, Epidemiology and End Results registries: does marriage affect cancer survival by gender and stage? Cancer Epidemiol. 35 (2011) 417–422. [16] J.L. Nelles, S.A. Joseph, B.R. Konety, The impact of marriage on bladder cancer mortality, Urol. Oncol. 27 (2009) 263–267.
23
[17] K. Saito-Nakaya, N. Nakaya, T. Akechi, M. Inagaki, M. Asai, K. Goto, et al., Marital status and non-small cell lung cancer survival: the lung cancer database project in Japan, Psychooncology 17 (2008) 869–876. [18] J.D. Sammon, M. Morgan, O. Djahangirian, Q.D. Trinh, M. Sun, K.R. Ghani, et al., Marital status: a gender-independent risk factor for poorer survival after radical cystectomy, BJU Int. 110 (2012) 1301–1309. [19] M.J. Rutherford, P.W. Dickman, P.C. Lambert, Comparison of methods for calculating relative survival in population-based studies, Cancer Epidemiol. 36 (2012) 16–21. [20] P.K. Andersen, R.B. Geskus, T. de Witte, H. Putter, Competing risks in epidemiology: possibilities and pitfalls, Int. J. Epidemiol. 41 (2012) 861–870. [21] S.R. Hinchliffe, P.C. Lambert, Flexible parametric modelling of cause-specific hazards to estimate cumulative incidence functions, BMC Med. Res. Methodol. 13 (2013) 13. [22] P. Dasgupta, D.R. Youlden, P.D. Baade, An analysis of competing mortality risks among colorectal cancer survivors in Queensland, 1996-2009, Cancer Causes Control 24 (2013) 897–909. [23] S.D. Fossa, M. Cvancarova, L.L. Chen, A.L. Allan, J. Oldenburg, D.R. Peterson, et al., Adverse prognostic factors for testicular cancer-specific survival: a population-based study of 27,948 patients, J. Clin. Oncol. 29 (2011) 963–970. [24] B.S. Rose, J.H. Jeong, S.K. Nath, S.M. Lu, L.K. Mell, Population-based study of competing mortality in head and neck cancer, J. Clin. Oncol. 29 (2011) 3503– 3509. [25] Queensland Cancer Registry, Cancer in Queensland: Incidence, Mortality, Survival and Prevalence, 1982 to 2012, QCR, Cancer Council Queensland and Queensland Health, Brisbane, 2014. [26] P.D. Baade, P. Royston, P.H. Youl, M.A. Weinstock, A. Geller, J.F. Aitken, Prognostic survival model for people diagnosed with invasive cutaneous melanoma, BMC Cancer 15 (2015) 27. [27] P.D. Baade, G. Turrell, J.F. Aitken, Geographic remoteness, area-level socioeconomic disadvantage and advanced breast cancer: a cross-sectional, multilevel study, J. Epidemiol. Community Health 65 (2011) 1037–1043. [28] P.D. Baade, P. Dasgupta, J.F. Aitken, G. Turrell, Distance to the closest radiotherapy facility and survival after a diagnosis of rectal cancer in Queensland, Med. J. Aust. 195 (2011) 350–354. [29] L.J. Krnjacki, P.D. Baade, B.M. Lynch, J.F. Aitken, Reliability of collecting colorectal cancer stage information from pathology reports and general practitioners in Queensland, Aust. N. Z. J. Public Health 32 (2008) 378–382. [30] Australian Bureau of Statistics. Marriages and Divorces, Australia, 2012 (cat. no. 3310.0). (http://www.abs.gov.au/ausstats/[email protected]/Products/ 3310.02012ChapterMarriages?OpenDocument) ABS, Canberra, 2012 (accessed 03.02.2015). [31] P.C. Lambert, P. Royston, Further development of flexible parametric models for survival analysis, Stata J. 9 (2009) 265–290. [32] P. Royston, P.C. Lambert, Flexible Parametric Survival Analysis Using Stata: Beyond the Cox Model., Stata Press, Texas, 2011. [33] M.A. Cleves, R.G. Gutierrez, M.W. Gould, Y.V. Marchenko, An Introduction to Survival Analysis using Stata, 3rd ed., Stata Press, 2010 Revised ed.. [34] J.P. Fine, R.J. Gray, A proportional hazards model for the subdistribution of a competing risk, J. Am. Stat. Assoc. 94 (1999) 496–509. [35] S.R. Hinchliffe, P.C. Lambert, Extending the flexible parametric survival model for competing risks, Stata J. 13 (2013) 344–355. [36] S. Eloranta, J. Adolfsson, P.C. Lambert, P. Stattin, O. Akre, T.M.L. Andersson, et al., How can we make cancer survival statistics more useful for patients and clinicians: an illustration using localized prostate cancer in Sweden, Cancer Causes Control 24 (2013) 505–515. [37] J. Giese-Davis, A. Waller, L.E. Carlson, S. Groff, L. Zhong, E. Neri, et al., Screening for distress, the 6th vital sign: common problems in cancer outpatients over one year in usual care: associations with marital status, sex, and age, BMC Cancer 12 (2012) 441.