Adherence to Recommended Treatments for Early Invasive Breast Cancer : Decisions of Women Attending Surgeons in the Breast Cancer Audit of Australia and New Zealand

Adherence to recommended treatments has been linked to higher disease-free and overall survival for breast and other cancers (Li et al., 2000; Herbert-Croteau et al., 2004; McCowan et al., 2008; Wockel et al., 2010a,b; Hershman et al., 2011). Non-adherence can be a result of many clinical factors including treatment compromises to accommodate patient frailty or co-morbidity, clinician preference, poor service access or high user cost (Ruddy et al., 2009; Lebeau et al., 2011). Non-adherence has also been linked to sub-optimal coordination across multidisciplinary teams, inadequate organization and delivery of services, low clinician case load, and cancer features such as size of lesion and nodal status (Bloom et


Introduction
Adherence to recommended treatments has been linked to higher disease-free and overall survival for breast and other cancers (Li et al., 2000;Herbert-Croteau et al., 2004;McCowan et al., 2008;Wockel et al., 2010a,b;Hershman et al., 2011).Non-adherence can be a result of many clinical factors including treatment compromises to accommodate patient frailty or co-morbidity, clinician preference, poor service access or high user cost (Ruddy et al., 2009;Lebeau et al., 2011).Non-adherence has also been linked to sub-optimal coordination across multidisciplinary teams, inadequate organization and delivery of services, low clinician case load, and cancer features such as size of lesion and nodal status (Bloom et & Partridge, 2009).
The Breast Surgeons' Society of Australia and New Zealand (formerly the Breast Section of the Royal Australasian College of Surgeons) aims to maintain and enhance high quality care and optimal treatment outcomes by promulgation of evidence-based treatment guidelines, continued medical education, and monitoring of practice and identification of statistical outliers (Wang et al., 2008a;2008b).There is concern that when women decline recommended therapies, technical quality may suffer and treatment outcomes may be jeopardized.Nonetheless it is recognized that the final choice of treatment is the patient's to make.
The aim of this study is to determine the extent to which patients decline recommended treatments, with a view to: (1) establishing benchmarks for future monitoring; and (2) considering policy and research implications.In this study, the frequency with which patients decline recommended treatments is investigated, overall and by treatment, age, cancer and health-system descriptor.Although reasons for declining recommendations are not recorded in the Audit, possible reasons and implications for policy development and research are discussed.

Materials and Methods
Audit data were collected by participating surgeons for an estimated 60% of early breast cancers diagnosed in Australia and New Zealand during the study period (Wang  et al., 2008 a, b ; Roder et al., 2011).Although breast cancers included in the Audit are not selected to be representative of all early breast cancers, their characteristics appear similar.The Audit database has shown, for example, similar survivals to those for early breast cancers recorded at a population level in New South Wales and the USA [Surveillance Epidemiology and End Results (SEER) data] (Roder et al., 2010(Roder et al., : 2011)).In addition, differences in survival by conventional risk factors, such as tumour size, grade, nodal status and oestrogen receptor status, have accorded with differences expected from population-based data, which has added credibility to the Audit database (Roder et al., 2010).
Breast Cancer Audit data have been collected by members of the Breast Surgeons' Society of Australia and  New Zealand since 1998 (Wang et al., 2008 a, b ).Data items include patient age, breast surgery type (i.e., mastectomy or complete local excision), axillary lymph node surgery, and referral for radiotherapy, chemotherapy, hormone therapy, and other systemic therapies.Patients' decisions to accept or decline treatments have been recorded.Data are collected on tumour histology, size, grade, lymphatic/ vascular invasion, hormone receptor status, and axillary nodal status, using a minimum data set with clear data dictionary definitions.Names of treatment centres have been collected plus women's private health insurance status.
Survival data are not collected from surgeons but are derived for Australian women through linkage with the National Death Index at the Australian Institute of Health and Welfare, using the first three digits of the surname, dates of birth and jurisdiction of residence for probabilistic matching (Roder et al., 2010).Linkage accuracy was investigated in a pilot investigation of data for 1,179 women treated by South Australian surgeons.Accuracy of linked death data was compared with death information recorded on the South Australian Cancer Registry for the same women where full names were available for linkage and resolution of doubtful links had been undertaken by Registry staff through active followup.The results showed a high accuracy, with a linkage sensitivity of breast-cancer death detection of 93.1%, a specificity of 99.9%, a predictive value positive of 96.4%, and a predictive value negative of 99.8% (Roder et al., 2010).
Following the pilot, death data were obtained from the National Death Index for all Australian women recorded on the Audit database.The date of censoring of live cases in the survival follow-up was December 31 st , 2007.In New Zealand, a similar follow-up of deaths was undertaken through the National Mortality Collection by deterministic matching using the National Health Index.This Index comprises a unique alphanumeric identifying number for each New Zealander.
Data for patients diagnosed in 1998-2005 were used in this study to allow enough follow-up time for survival assessment.A total of 36,755 cases and 2,410 breast cancer deaths were included.They excluded a small number for whom data were unavailable on whether treatment had been declined.The mean follow-up period from diagnosis was 58 months, with a range from less than one to 119 months.
Cross-tabulations were first undertaken to produce rate ratios (95% confidence limits) for declining a treatment (Garlinger & Abramson, 1995), both for treatment of any type collectively, and for individual treatment types, and by age, tumour characteristic, patient insurance status, annual case load of surgeon (i.e., ≤20, 21-100, >100), and location of treatment centre (major city, inner regional and more remote, using the Australian Standard Geographic Classification (AIHW et al., 2008) and equivalent geographic groupings for New Zealand).Statistical testing of associations with treatment adherence was undertaken using the Pearson chi-square test for binary and nominal variables and the Mann-Whitney U test for ordinal variables (Armitage & Berry, 1987;StataCorp, 2005).
Multiple logistic regression was used to confirm findings from these cross-tabulations, with a separate analysis undertaken for each treatment type.Declining treatment (yes/no) was the dependent variable in these analyses and explanatory variables were those which achieved statistical significance (p<0.05) in backwards elimination.Underlying assumptions such as lack of colinearity were tested and found to be met (Armitage & Berry, 1987;StataCorp, 2005).
Although only 81 breast cancer deaths were available for analysis among women declining treatment, which substantially reduced statistical power.Relative risks of case fatality (i.e., hazards ratios) and 95% confidence intervals were calculated using Cox proportional hazards regression (Armitage & Berry, 1987;StataCorp, 2005).Survival times were calculated from diagnosis to December 31st, 2007 or date of death, whichever occurred first.Variables entered as predictors included declining treatment (yes/no), and as co-variables, age at diagnosis and clinical risk factors (i.e., histology type, tumour size, grade, nodal status, lymphatic/vascular invasion and oestrogen receptor status) categorized as shown in Tables 1  and 2. Proportionality and lack of co-linearity assumptions were checked and found to be satisfied.
The Cox models can be regarded as multivariate equivalents of disease-specific survival analyses.Diseasespecific survivals have been shown in population-based studies in Australia to correspond closely with relative survival (Armitage & Berry, 1987;SACR 2000SACR , 2007)).Disease-specific survivals are often preferred in clinical studies where, due to referral practices, patients may not have risks of competing causes of death equivalent to population norms (a required assumption for relative survival).

Characteristics associated with declining treatment
Bivariate analyses: associations of person and health system descriptors with declining treatment are shown in Table 1.Univariate predictors of declining a treatment included: • Age at diagnosis (p<0.001):The percentage declining a treatment increased from 2.6% in women under 40 years to 5.8% in those aged 80 years or more.Declining a treatment increased with age for axillary surgery (p=0.006),radiotherapy (p<0.001), and mastectomy (for women aged 80 years or more) (p<0.001).By comparison, age was not associated with declining breast conserving surgery (p=0.131),chemotherapy (p=0.809),hormone therapy (p=0.107) or multiple therapies (p=0.168).
• Annual surgeon case load (p<0.001):The percentage declining a treatment was higher for lower case loads, ranging from 2.6% where the case load exceeded 100 cases per year to 6.3% where the case load was ≤ 20 cases per year.Declining treatment was higher for lower case load for breast conserving surgery (p<0.001),mastectomy (p<0.001),axillary surgery (p<0.001),radiotherapy (p<0.001),chemotherapy (p=0.015), and multiple treatments (p=0.008).No association was evident for hormone therapy (p=0.648).
• Private health insurance: Although not statistically significant (p=0.070),declining a treatment tended to be more common among women without private health insurance, due to differences for declining mastectomy (p=0.014) and multiple treatments (p<0.001).
The only tumour descriptor associated with declining a treatment as a univariate predictor was low or intermediate as opposed to high grade (p=0.011)(Table 2).The percentage declining a treatment was 3.6% for low, 3.4% for intermediate and 3.0% for high grade, due to differences for declining axillary surgery (p=0.024),hormone therapy (p<0.001), and possibly radiotherapy (p=0.053).By comparison a reverse trend was indicated for declining chemotherapy (p<0.001).
• Grade -compared with low grade, 0.96 (0.83, 1.10) for intermediate grade and 0.84 (0.71, 0.98) for high grade.Private health insurance status was not associated with declining a treatment (p>0.200).

Characteristics associated with declining specific treatment types
• Breast conserving treatment: Descriptors associated with declining breast conserving treatment in bivariate analyses included: low case load (p<0.001);inner regional treatment location (p<0.001); and fewer cancer foci (p=0.010).Multiple regression analysis confirmed that declining this treatment was related to case load (p<0.001),inner regional treatment location (p<0.001) and fewer breast cancer foci (p=0.042).Odds ratios were: for annual case load -compared with up to 20 cases, 0.19 (0.13, 0.27) for 21-100 cases and 0.06 (0.03, 0.14) for over 100 cases; for treatment location -compared with major city, 2.03 (1.38, 2.99) for inner regional and 0.45 (0.11, 1.86) for more remote areas; and for number of cancer focicompared with one focus, 0.62 (0.27, 1.43) for two foci and 0.35 (0.13, 0.97) for three or more foci.

Survival
The number of breast cancer deaths available for analysis was small among women declining treatment (n=81) and smaller again by sub-category of treatment type declined.As a consequence, statistical power for detecting survival differences was very low.Nonetheless, after adjusting for age and conventional clinical risk factors, the relative risk (hazards ratio) was: for breast surgery (mastectomy and breast conserving surgery)compared with surgery received, 2.29 (1.86, 2.82) when it was not recorded to have been offered, and 1.25 (0.67, 2.36) when it was declined; for radiotherapy -compared with radiotherapy received, 1.22 (0.63, 2.35) when it was declined; and for systemic therapy, compared with systemic therapy received, 1.11 (0.57, 2.13) when it was declined (note: the corresponding relative risk for declining chemotherapy was 1.41 (0.83, 2.40)).An analysis stratified by treatment type gave an unadjusted relative risk of 2.40 (1.27, 3.53) for declining treatment and a corresponding adjusted figure of 1.19 (0.71, 1.67).

Discussion
Only 3.4% of women were reported to have declined a recommended treatment, indicating that clinician recommendations were largely followed.While reasons for declining treatments were not recorded, differences in declining treatments were investigated by age, treatment location and other health-system descriptors, which generally accorded with results of other studies (Bloom et al., 2004;Truong et al., 2004;Gilligan et al., 2007;Wockel et al., 2010a ;Hershman et al., 2011;Lebeau et al., 2011;Weggelaar et al., 2011).
These results provide a basis for hypothesis generation, particular when supported by other study results.For example, consistent with results of prior studies of non-adherence, the percentage of women declining different treatments was higher for chemotherapy than surgery (Wockel et al., 2010 a ).This may reflect the lesser importance attributed by women to adjuvant therapy and a desire to avoid side effects.
Low surgeon case load has also been found previously to relate to non-adherence to recommended clinical guidelines (Gilligan et al., 2007).In the present study, the percentages of women declining breast conserving surgery, mastectomy, axillary surgery, radiotherapy and chemotherapy, respectively, were higher where the surgeon case load was lower.This applied in multivariable analysis after adjusting for age, cancer and health-system descriptors.The reasons are unknown but may relate to multidisciplinary care.A report from the National Breast and Ovarian Cancer Centre found a negative association of low case load with provision of multidisciplinary care (NBOCC, 2008).Qualitative research should be directed at low case-load settings to explore reasons and promotional strategies.
In general, older women were more likely to decline a treatment recommendation, which is consistent with non-adherence results from other studies (Truong et al.,  2004; Schaapveld et al., 2005; Wockel et al., 2010 a ; Lebau  et al., 2011; Weggelaar et al., 2011).Declining treatment recommendations was most apparent for radiotherapy.This may reflect the duration of this care (often about six weeks) and access issues due to the more limited numbers of treatment outlets available.Older women may be deterred from seeking treatments under these circumstances, due to effects of increasing age-related frailty and co-morbidity on physical capacity and selfconfidence.Given their stage of life, these women may be more inclined to forego treatment, particularly radiotherapy and other treatments seen as having an adjuvant role.
More recently, the National Breast and Ovarian Cancer Centre developed guidelines for hypofractionation radiotherapy techniques in breast cancer that older women may find more convenient and may lead to stronger adherence to radiotherapy recommendations (CA, 2011).Furthermore, abbreviated partial breast radiotherapy may be a suitable option for some older women with breast cancer undergoing breast conserving surgery.The availability of this treatment may also lead to stronger adherence to radiotherapy recommendations.
Remoteness is well-known to reduce opportunities to gain recommended treatments (Li et al., 2000;BCNA, 2011).This was also demonstrated in the present study where women treated in more remote areas were more likely to decline a treatment recommendation.This applied in particular for radiotherapy, where travel and accommodation requirements would have been greatest.The use of hypofractionation and abbreviated partial breast irradiation may also assist in increasing adherence to radiotherapy recommendations by this sub-group of women.Remoteness of location was also demonstrated to reduce adherence for hormone therapy recommendations.The reason for this is not clear.
Tumour grade has been linked to non-adherence in a number of studies, although results have been inconsistent (Lebau et al., 2011).In the present study, women with lowgrade tumours were more likely to decline recommended axillary surgery and hormone therapy.Again, the reasons are unclear, although it may be that low grade tumours would appear less life-threatening and women may feel that they have more latitude to decline recommended treatments.
Other more specific findings included a greater tendency to decline a recommended mastectomy in the absence of private health insurance and when axillary nodal status was positive.This may reflect variations in access to treatment through the public and private hospital system and potentially, a perspective that mastectomies are less likely to clear cancers that have already spread beyond the breast.Meanwhile the greater likelihood of refusing chemotherapy when tumour sizes were larger, and hormone therapy when diameters were smaller, nodal status was negative, and the histology type was ductal, is unexplained.
Priority ideally is given to randomized trial evidence in the development of clinical guidelines.However evidence from observational research can also be of value.Cohorts of women who have declined treatments provide an opportunity to investigate effects of treatment shortfalls on multiple outcomes, including daily function and quality of life, and broader effects on families, in addition to recurrence-free and overall survival.These and other research opportunities should be pursued.
Non-adherence to treatment recommendations have been shown to correlate with lower cancer survivals in previous studies cancers (Li et al., 2000; Herbert-Croteau  et al., 2004; McCowan et al., 2008; Wockel et al., 2010 a,  b ; Hershman et al., 2011).While the statistical power DOI:http://dx.doi.org/10.7314/APJCP.2012.13.4.1675 Adherence to Recommended Treatments for Early Invasive Breast Cancer in Australia andNew Zealand available to demonstrate similar associations in the present study was very low due to small numbers, point estimates were suggestive of higher case fatalities when treatments were declined.
It is important that further research be undertaken to determine reasons for declining treatments.These should include qualitative studies that have greater flexibility to explore differences in knowledge, attitudes, perceived risks, barriers to care, and perhaps other patient factors such as ethnicity and educational status.
In conclusions, women justifiably have the right to choose their treatment and can elect to decline clinician recommendations.Reasons for declining recommendations require further investigation however, to ensure that choices are well informed and clinical outcomes are optimized.

Table 1 . Rate Ratios (95% Confidence Limits) of Women Declining Clinician Recommendations for Breast Cancer Treatment by Age and Health-System Characteristics; BreastSurgANZ National Breast Cancer Audit, 1998-2005
*MW = Mann-Whitney U test; X 2

Table 4 . Relative Risks (95% Confidence Limits) of Breast Cancer Death According to Whether Recommended Treatments were Declined, by Treatment Type; BreastSurgANZ National Breast Cancer Audit, 1998-2005*
Dec 31 st , 2007, **Adjusted for age and diameter, grade, nodal status, oestrogen receptor status, vascular invasion, and number of cancer foci (see Tables1 & 2), categories included for unknown values.
*Cox proportional hazards regression; date of censoring of live cases =