Study population

This study involved all women who were diagnosed with primary breast cancer in the Auckland and Waikato District Health Board Regions, where about two-fifths of the country’s population reside and which have similar incidence and mortality rates from breast cancer compared to the national figures [2], between June 2000 and May 2013.

Data sources

The participants were identified from the Auckland and Waikato Breast Cancer Registers which are prospectively maintained population-based databases. From 2000 onward, the registers capture almost all newly diagnosed breast cancer cases in the respective district health board regions. The completeness of the Waikato Breast Cancer Register has been checked against the National Cancer Registry, and found to be 99% complete. The Auckland Breast Cancer Register is also complete since the removal of consent requirement in 2012 with 1% lost to follow up. Both databases contain more comprehensive and accurate information compared with national data sources [21–23].

Using the National Health Index (NHI) number, a unique identifier assigned to every person who uses health and disability support services in New Zealand, the registers are regularly linked to the National Cancer Registry and Mortality Collection. The New Zealand Cancer Registry contains information about all malignant tumours first diagnosed in New Zealand, except basal cell and squamous cell tumours of the skin. The Mortality Collection contains information about all deaths registered in the country [24]. The data were also linked to the National Minimum Dataset to obtain information on comorbidities. The National Minimum Dataset contains information about all day patients and inpatients discharged from all public hospitals and over 90% of private hospitals in New Zealand [25].

Variables of interest

The exposure of interest in this analysis was the type of health care facility (public vs. private) where breast cancer was primarily treated (mostly surgical treatment), and the primary study outcome was breast cancer specific mortality. The categorisation of the death being due to breast cancer was made based on the medical records and death certificates. Information on the cause of death was ascertained by referring to original documents and cross-referencing with other national databases.

Other variables were selected based on their likely confounding or mediating effect on the exposure-outcome association, and include: patients’ demographics such as age, ethnicity and health domicile code, year of cancer diagnosis, mode of presentation (screen or symptomatic), tumour characteristics such as stage at diagnosis (Tumour, Node and Metastasis (TNM) system), grade, histological type and hormone receptor status, and treatments undertaken such as time to first treatment, loco-regional therapy (i.e., surgery and radiotherapy), chemotherapy and hormonal therapy. The health domicile codes represent patients’ usual residential address and categorised as urban (main urban, satellite urban and rural with high urban influence) and rural areas (others) based on Statistics New Zealand’s Urban/Rural Profile [26]. To assess the degree of neighbourhood deprivation, the domicile codes were also mapped on to the 2006 New Zealand Deprivation Index (NZDep) [27] with decile ten the most deprived and decile one the least. Patients’ comorbidity was measured using a C3 index score which is a cancer-specific index of comorbidity based on the presence of 42 chronic conditions recorded in the National Minimum Dataset for a period of five years prior to the diagnosis of cancer [28]. Each condition is weighed to its impact on one-year non-cancer mortality in a cancer cohort, and the weights are then summed to get a final comorbidity score.

Analyses

Baseline data were presented as means with standard deviations and medians with interquartile ranges for continuous variables and percentages for categorical variables. All the data were complete for only 8505 participants (59.3%) as a large number of records, particularly prior to 2006, had missing data relating to HER-2 status. If HER-2 status was excluded, the data were complete for 75.0% of the participants. Missing values were computed using multiple imputation with ten complete datasets created by the Markov chain Monte Carlo method [29], incorporating all baseline co-variables and survival outcomes. Differences in baseline characteristics between patients who received public vs. private care were assessed using a two-sample T-test (for continuous variables) and Chi-squared test (for categorical variables) and adjusted for the year of diagnosis using PROC MIXED which fits a variety of mixed linear models to the data. Differences in the types of cancer treatment received were also adjusted for stage at diagnosis and biological factors.

To assess hazards of breast cancer specific mortality associated with the type of health care received, Cox proportional hazards regression modelling was performed with death from breast cancer as the failure variable, and death from another cause, or if alive, date of last follow-up, as censored observations. Hazard ratios (HR) were sequentially adjusted for five domains of co-variables: demographics, mode of presentation, disease factors, comorbidity index and treatment factors. When the continuous variables (age, time to first treatment and C3 index score) were added to the model, restricted cubic splines were used with knots at the 5th, 50th, and 95th percentiles [30]. Both age and menopausal status were retained in the models as R² was 0.54 (equivalent to a variance inflation factor of 2.17) when the former was regressed on the latter. HER2 status was excluded as about one-third of the records had missing values and its impact on the exposure-outcome association was negligible.

The mediating role of each domain was determined by the percentage reduction in the β coefficient after inclusion of each domain in the model using the approach described previously [31]: 100 × (β_crude-β_adjusted)/β_crude. The 95% confidence intervals relating to each percentage attenuation were estimated using a nonparametric bootstrapping method with 2000 re-samplings (with replacement). Subgroup analyses were undertaken by hormone receptor status. Sensitivity analysis was performed using total mortality (deaths from any cause) as the outcome variable. All analyses were performed using SAS (release 9.4, SAS Institute Inc., Cary, North Carolina).

Ethics statement

This study extracted data on diagnosis, treatment and outcomes of patients diagnosed with breast cancer from the two registers, and linked the data to other national data collections such as the National Minimum Dataset and Mortality Collection. Written informed consent was not sought as it was not feasible to trace all patients. The data were analysed anonymously. Ethical approval for this study was obtained from the New Zealand Northern ‘A’ Ethics Committee (Ref. No. 12/NTA/42).

Main findings

In this study, about 62% of patients received public health care and they had a 95% higher average risk of mortality compared to those who received private care. There were significant differences in demographics, mode of presentation, disease factors, comorbidity index and treatment factors between the two populations, which explained approximately 80% of the survival disparities. The three key contributors were ethnicity, stage at diagnosis and type of loco-regional therapy.

Strengths and limitations

This study used the data from two prospectively maintained population-based databases which contain comprehensive and near complete information about patients diagnosed with primary breast cancer. Linkage to the national databases also enabled us to ascertain information on cause of death and to obtain information on comorbidities. Yet, a considerable proportion of patients had missing data particularly relating to HER-2 status, most of whom were diagnosed prior to 2006 when HER-2 testing was not routine in New Zealand. We excluded this variable from the analyses as its impact on the exposure-outcome association was negligible. It was also not possible to assess the impact of some important factors such as smoking as such information was not recorded in the databases. NZDep2006 used in this analysis measures area-level deprivation and may not reflect an individual’s actual socioeconomic status although it has been validated previously [32]. Additionally, some patients treated in the private sector may have been transferred to a public facility for further care after their primary treatment. This may result in misclassification of exposure and bias the association estimates. Potential misclassification of cancer-specific deaths may also occur but such errors are likely to be similar in any subgroups being compared, and will only act to reduce observed differences to a small extent.

Interpretations

A higher risk of mortality in patients who received public care can be partly explained by demographic differences in health care access. Overseas studies reported that access to private care was less common in ethnic minorities and socioeconomically disadvantaged groups [9, 12–16]. Similar findings were observed in this study. In particular, ethnic variations appear to be a major contributor to survival disparities. It is well documented that Māori and Pacifika women have poorer survival from breast cancer compared to other ethnic groups, which is mostly contributed by differential access to cancer care services [5, 33–35]. In a recent population-based case-control study involving about 1800 patients, a higher proportion of Maori and Pacifika women reported barriers to and delays in access to care compared to other ethnic groups with the most commonly reported barriers being “cost” and “fear” [36]. This underscores the need for more efforts to improve affordable and culturally sensitive cancer care services for Maori and Pacifika patients. Such efforts should be focused on the quality and timeliness of public health services where these patients are more likely to be treated.

After adjusting for demographic factors, patients in the public sector still had a 56% higher risk of mortality from breast cancer compared to those in the private sector. This could be due to deficits in the public health care system along the cancer care pathway. Consistent with previous research [8–18], our analysis found that patients who received public care were less likely to be diagnosed with early staged cancer and to receive timely cancer treatments.

Stage at diagnosis is an indicator of health care access and is a key contributor to survival disparities [33, 37]. Likewise in this study, late diagnosis has contributed significantly to survival differences between private and public care, and could be partly explained by differential access to screening services–the proportion of patients with screen-detected breast cancer was lower in the public sector than the private sector. New Zealand has a national breast cancer screening program established in 1998, which initially offered publicly funded mammography to all asymptomatic women aged 50 to 64 years and was extended to include women aged 45 to 49 years and 65 to 69 years in 2004. While screening coverage has improved over time and met the target of 70%, coverage for Māori women is relatively low [38], and there is room for improvement. Other factors that could contribute to late diagnosis include patient factors such as health literacy, low socioeconomic status, fear of cancer and other psychosocial factors, and health system factors such as those related to primary care practices and practitioners [39–42], and are worthy of further investigation.

In addition to experiencing diagnosis delays, patients treated in the public sector had a significantly longer time (40 more days on average) to the first treatment after diagnosis as reported previously [43]; however, treatment delays have not accounted substantially for their inferior survival. A more important contributing factor is differences in loco-regional therapy received. While some patients in the public sector may have been treated by less experienced surgeons with lower case loads, which has been associated with poorer outcomes in previous overseas studies [44, 45], omission of definitive treatment is much more likely to be the cause. We found that patients in the public sector were less likely to receive primary surgery or radiotherapy. This is not surprising as the private sector provides quicker access to elective surgery and other health services which are mostly funded by health insurance. We also found a lower rate of breast conserving surgery in patients treated in the public sector as reported in some US studies [11, 46]. This may reflect a higher prevalence of advanced cancer in the public sector but our analysis accounted for stage at diagnosis and biological factors. Another possible explanation is patient preference. A previous Australian study reported that patients in the public sector were less likely to accept the recommended standard chemotherapy [47] but little is known with regard to loco-regional treatments in breast cancer patients.

In New Zealand, some initiatives are in place to improve cancer care and support for patients and their families, for example, the Faster Cancer Treatment (FCT) program established by the Ministry of Health [48]. However, survival disparities observed in this analysis as well as in previous research suggest that a lot more needs to be done to improve the quality, timeliness and equitability of public cancer care services.