Impact of deviation from guideline recommended treatment on breast cancer survival in Asia

Breast cancer survival has improved with significant progress in treatment and disease management. However, compliance with treatment varies. Treatment guidelines for older patients are unclear. We aim to identify predictors of noncompliance with recommended therapy in a large breast cancer population and assess the impact of noncompliance on survival. Our study included 19,241 non-metastatic female breast cancer patients, of whom 3,158 (16%) died within 10 years post-diagnosis (median survival = 5.8 years). We studied the association between treatment noncompliance and factors with logistic regression, and the impact of treatment noncompliance on survival with a flexible parametric survival model framework. The highest proportion of noncompliance was observed for chemotherapy (18%). Predictors of noncompliance with chemotherapy, radiotherapy and endocrine therapy included age, tumor size, nodal involvement and subtype (except radiotherapy). Factors associated with not receiving surgery included age and subtype. Treatment noncompliance was associated with worse overall survival for surgery (HR: 2.26 [1.80–2.83]), chemotherapy (1.25 [1.11–1.41]), radiotherapy (2.28 [1.94–2.69]) and endocrine therapy (1.70 [1.41–2.04]). Worse survival was similarly observed in older patients for whom guidelines generally do not apply. Our results highlight the importance of following appropriate treatment as recommended by current guidelines. Older patients may benefit from similar recommendations.


Study population.
Hospital-based registry data on 20,999 citizens and permanent residents of Singapore diagnosed with breast cancer (ICD 9: 174*; ICD 10: C50*) between 2005 and 2015 was obtained from six major restructured public hospitals in Singapore (Changi General Hospital [CGH], KK Women's and Children's Hospital [KKH], National Cancer Center Singapore [NCCS], National University Hospital [NUH], Singapore General Hospital [SGH], and Tan Tock Seng Hospital [TTSH]). Patients who were diagnosed with metastatic breast cancer (n = 1,644), who were male (n = 26), whose date of birth was not reported (n = 6), or who had missing information on death date (n = 82) were excluded (Fig. S1). Ethical approval for using the de-identified pooled data was obtained from the National Healthcare Group Domain Specific Review Board (NHG DSRB Ref: 2013/01085 and CIRB Ref: 2016/3010). The institutional review boards of the respective hospitals involved in this study had approved the conduct of this study without the need for informed consent.
Compliance with treatment guidelines. The restructured public hospitals in Singapore generally follow the guidelines of the NCCN and the St Gallen 2005 consensus (Table 1) 9,20 . For each treatment (chemotherapy, radiotherapy and endocrine therapy), patients were classified into three groups: (1) started recommended treatment [compliant], (2) did not start recommended treatment [noncompliant], or (3) treatment not recommended [compliant]. All patients were non-metastatic at diagnosis and should be recommended surgery (i.e. started recommended treatment [compliant] or did not start recommended treatment [noncompliant]).
Demographic and clinical characteristics. Demographic   Outcome of interest. The primary outcome of interest was overall survival, with time since diagnosis in years as the underlying time scale. Patients were followed up until death, or censored due to either loss of follow-up or end of study period (ten years after date of diagnosis Statistical analysis. Bivariate associations between vital status (dead or alive) and patient-related characteristics (demographic, clinical and treatment) were evaluated using the Chi-square test. To identify patient-related characteristics associated with the compliance with recommended treatment, logistic regression was used to estimate the odds ratios (OR) and corresponding 95% confidence intervals.
To examine the effect of treatment compliance on survival, mortality rate was calculated. This risk estimate was defined as the number of deaths divided by risk time (total person-years [PY] for which the patients were alive), modeled using flexible parametric survival model, with cubic splines function for baseline mortality. The associations between individual treatments and mortality rates (unadjusted) was first modelled assuming proportion hazards throughout the follow-up time. A flexible parametric survival model framework was then used to model time-dependent effects (STATA version 8, command: stpm2), hazard ratios (HR) and corresponding 95% confidence intervals were estimated. The Akaike Information Criterion (AIC) statistic was used to optimize the number of internal knots (i.e. degrees of freedom of the baseline model) and
Mortality rates by treatment compliance. Not undergoing recommended surgery resulted in worse survival, particularly in the first year after breast cancer diagnosis (Fig. 1a). A larger proportion of patients who did not start recommended chemotherapy or radiotherapy died compared to patients who started recommended treatment and patients for whom treatment was not recommended (Fig. 1b,c). While mortality rates for patients who were not recommended treatment (vs. started treatment) were slightly lower for chemotherapy and radiotherapy, the mortality rate for patients who were not recommended endocrine therapy (i.e. patients who are ER-and PR-negative) was higher than the corresponding rate for patients who started endocrine therapy within ~6 years post-diagnosis (Fig. 1d).

Discussion
Treatment compliance, which has an impact on survival, was high for primary surgery, radiotherapy and endocrine therapy but low for chemotherapy. Factors associated with chemotherapy, radiotherapy and endocrine therapy noncompliance included older age, smaller pre-surgical tumor size, and lower nodal stage. Patients with luminal A proxy subtype were more likely to be noncompliant with chemotherapy than other proxy subtypes. Patients with luminal B [HER2-positive] or luminal B [HER2-negative] proxy subtype were more likely to be noncompliant with endocrine therapy than patients with luminal A proxy subtype. Noncompliance with recommended treatment (chemotherapy, radiotherapy, or endocrine therapy) was associated with a ~1.5 to 6-fold increased risk of dying within 10 years post-diagnosis. In patients aged ≥70 years who were recommended chemotherapy, noncompliance with recommended chemotherapy and radiotherapy were associated with worse survival.
Factors associated with treatment noncompliance. In agreement with previous studies, tumor characteristics (pre-surgical tumor size, grade, proxy subtype) and age at diagnosis were associated with initiation of recommended treatment [22][23][24] . Many factors can potentially influence a patient's decision to go along with recommended treatment: previous experiences and personal values 17 , fear of treatment-related side effects 18 , financial ability, age, education and health status, among others 19,20 . Although information on financial status was not available, the subjects in our study were patients at restructured hospitals and were eligible for subsidized treatment. It is thus less likely that any ethnic disparity observed was due to differences in financial resources 25 . An ethnic disparity in recommended treatment uptake has also been observed by others; recommended treatment uptake is frequently lower in minority populations 26,27 . Although higher noncompliance with chemotherapy and radiotherapy were observed in Indian patients as compared with Chinese patients, the observed worse Surgery*, n = 19,241 Chemotherapy*, n = 11,034 Radiotherapy*, n = 9,671 Endocrine therapy*, n = 12,251  Table 3. The odds of not receiving treatment with respect to demographic characteristics and treatment was studied using logistic model. *Women who started treatment was used as the reference category of the outcome, hence an odds ratio (OR) value above 1 implies that the exposure category (e.g. Indian) has higher odds of not starting recommended treatment than the reference level of the exposure factor (e.g. Chinese). CI: Confidence interval. ^Additionally adjusted for site. after accounting for compliance with treatment, pre-surgical tumor size, nodal status, age at diagnosis, year of diagnosis and center. Since Malay patients were just as likely to start recommended treatment as Chinese patients, it is unlikely that the survival difference was due to treatment compliance. Moreover, according to the work of Hill et al. in breast cancer patients in the United States, the disparity in the survival of different ethnicities among the older patients in Asia was not due to deviation from guideline recommended treatment 12 . Further studies to identify other factors that delay or prevent the timely receipt of recommended treatment would allow efforts to be directed towards their mitigation. Treatment compliance and overall survival. The high mortality rate associated with not proceeding with recommended surgery is in agreement with results reported in existing literature 10,13 . Notably, patients who did not follow recommended chemotherapy, radiotherapy or endocrine therapy were ~2 to 3 times more likely to die from breast cancer than patients who do 12-15 . Our results are in agreement with previous studies which found increased risk of death in patients who did not start endocrine therapy 12,15 . Interestingly, while mortality rates of breast cancer patients who followed recommended endocrine therapy were lower than that of patients who did not, the curves converged and crossed at ~2.5 and ~6 years after diagnosis (i.e. mortality rate for treatment compliant patients increased and higher than non-compliant patients). Endocrine therapy is typically recommended for at least five years. While we were not able to study drug adherence in our patients, unsatisfactory adherence to treatment for the entire recommended duration is known to be associated with worse survival, which could potentially explain the observed increasing mortality rate among treatment-compliant patients in our study 28 . As such, the survival benefit of endocrine therapy compliance would better reach its full potential if common factors contributing to non-adherence, such as patient forgetfulness and concerns about adverse effects, were followed up and addressed 29 .
Chemotherapy may appear to be less effective among patients with less aggressive breast cancers, where extensive treatment was unnecessarily administered for early stage disease [30][31][32] . For example, in patients who have favorable tumor characteristics (ER-positive/PR-positive, HER2-negative, and without nodal involvement) who met established guidelines for the recommendation of chemotherapy, those with favorable gene-expression profile may benefit from endocrine therapy alone 33 . Herr et al., found that chemotherapy did not improve survival  www.nature.com/scientificreports www.nature.com/scientificreports/ in patients with luminal A subtype, who had tumor size >20 mm and >3 positive nodes 34 . We observed that the survival benefit of starting recommended chemotherapy was more pronounced in patients with luminal A and basal proxy subtypes and starting recommended radiotherapy has the largest survival benefit in luminal B [HER2-positive] patients. We were not able to study comorbidities in our population which may have contributed to the contradictory findings with Herr et al. in luminal A subtype patients, as patients who are more frail are more likely to not start chemotherapy and also have worse survival.
Age at diagnosis and recommended chemotherapy. The NCCN guidelines restrict their recommendations on chemotherapy to breast cancer patients aged <70 years 20 . In addition, clinical trials on the use of chemotherapy typically do not recruit patients older than 70 years, providing little empirical support for the use of such  www.nature.com/scientificreports www.nature.com/scientificreports/ treatment in older patients 35 . Clinicians are less likely to recommend aggressive or invasive treatment such as chemotherapy or surgery for older women due to the increased likelihood of existing comorbidities 23,24 . Nonetheless, older patients can benefit from optimal anti-cancer therapy. For example, a previous study in stage I and II breast cancer patients found that having surgery reduced the risk of breast cancer specific death in patients aged ≥80 years 36 .
While chemotherapy compliance is generally associated with better survival, this was not the case for older patients who were noncompliant with radiotherapy. The reason for not starting recommended treatment is not known. Clinicians may at their discretion not advise guideline-recommended treatment for patients who were already in ill health. Patient level information such as comorbidities and performance status was not available as a proxy of health status at diagnosis. Hence, it is unclear whether the worse survival observed in patients who did not receive treatment was due to the lack of treatment or other causes (e.g. patients' preference, fear of adverse effects, or the lack of support) 26,37 .
Older patients may also be more prone to side effects from treatment, such as chemotherapy-induced neutropenia, fever and infection 38 . In older patients who complied with radiotherapy, chemotherapy conferred a substantial survival advantage. While the better survival linked to treatment compliance in the older breast cancer population could be an artifact due to more favorable health status, a similar survival benefit was still seen in a subset of patients who survived at least two years after diagnosis. This suggests that with close monitoring of patient response and the use of prophylactic agents such as granulocyte colony-stimulating factor to prevent adverse side effects, older patients too, may benefit from appropriate chemotherapy treatment 39 . Nonetheless, this result is noteworthy because women over 70 years of age comprise 13% of all breast cancer patients. Mortality rate is also highest for this age group.
Our study is not without limitations. Our classification of treatment was largely based on international guidelines which changes over time. However, most treatment modality and indication has seen small changes within our study period 2005-2015 40 Table 4. The introduction of targeted therapy, which is commonly administered to HER2 positive patients on chemotherapy, during this period (2005-2015) could not be accounted for. This may lead to an overestimation of the benefit of chemotherapy in our study. In an attempt to mitigate this influence, we performed stratified analysis by proxy subtype. We were not able to account for potential survival differences due to disparity in treatment intensity and toxicity. However, our result is likely to be biased towards the null, as the premature stopping of treatment, dose reduction and adverse drug reactions will reduce survival in the group of patients who started treatment. In addition, the evaluation of endocrine therapy impact on survival can be improved by studying the 5 year adherence of treatment instead of only the start of treatment. Information on comorbidities was not available, which may have result in contraindication of treatments and worse survival in patients who did not start treatment. Here, we partially accounted for comorbidities in patients aged >70 who would be more affected by comorbidities. There was no appreciable difference in the results (apart from uptake in surgery), when we excluded those who survived >2 years post-diagnosis in patients aged >70. The socioeconomic status and health seeking behavior of patients attending restructured hospitals (patients are eligible for subsidized care) may be different from patients attending private hospitals. However, it is recognized that the majority (~80%) of Singaporean's attend restructured hospitals and healthcare policies are targeted at these patients. Being geographically small with an extensive public transport system, refusal of recommended treatment due to inaccessibility is less common than in areas where specialized centers may be far from residential districts 41 . conclusion Noncompliance with recommended treatment was associated with worse survival (HR between 1.5 and 6.4). A survival benefit was observed when older breast cancer patients were treated with chemotherapy otherwise recommended only for younger patients (<70 years), suggesting that older patients may benefit from similar recommendations. Proxy subtype, pre-surgical tumor size, nodal stage, age at diagnosis and year of diagnosis were associated with treatment compliance. Our results highlight the importance of following appropriate treatment as recommended by current guidelines.

Data availability
Data used in this study is de-identified. Data is available upon request from Ms Yen Shing Yeoh (email: yen_ shing_yeoh@nuhs.edu.sg).