Australian researchers have developed a prognostic model that can estimate breast cancer-specific survival from the time of diagnosis of invasive breast cancer.
The model, developed by a team at Cancer Council Queensland, confirms key factors such as advanced stage at diagnosis, higher tumour grade and “triple negative” breast cancers as being predictive of poorer survival.
Based on long term follow up data from more than 3000 women diagnosed with breast cancer between 2010 and 2013, the results also highlight an independent survival benefit of breast cancers diagnosed through screening rather than being symptom detected.
The authors noted there was large variability in survival after a breast cancer diagnosis and the results showed that the final prognostic model explained about 36% of this variation, with ‘stage at diagnosis’ alone explaining 26% of the variation.
In a paper published in Breast Cancer Research and Treatment (link), the authors said previous survival models for breast cancer such as the Nottingham Prognostic Index (NPI) had performed well but were limited by the small number of variables available from cancer registry data.
They therefore aimed to develop a model based on information from the Breast Cancer Outcomes Study (BCOS), which involved 3323 patients, and supplemented with cancer registry data. The median follow-up time of the study was seven years and 251 women (7.6%) had died by the time of last follow-up, with 69% of the deaths attributed to breast cancer.
The initial list of eight variables related to breast cancer survival included age and stage at diagnosis, tumour size, grade and clinical subtype, positive lymph nodes, mode of detection and diagnostic delay.
A final model was based on age, stage at diagnosis, tumour grade, clinical subtype, and mode of detection. This had a strong predictive ability for breast cancer specific survival (Harrell’s C-statistic 0.84) while the area under the ROC curve for 5-year mortality was 0.87.
The model was then applied to predict the one-year and five-year survival probabilities for 12 hypothetical patients.
Overall, it tended to predict poorer survival among patients with advanced stage of disease and of older age, and higher survival for patients whose breast cancer was detected via screening rather than from symptoms and who had tumours of early stage and low grade.
Patients with tumours of Luminal A like clinical subtype had a better prognosis, while those with Triple Negative subtype tended to fare the worst.
For example, a 55-year-old patient with stage 1 tumour of low grade and Triple Negative subtype detected by screening had a predicted survival of 86.3%, while a patient of similar age with symptom detected advanced stage breast cancer of high grade and Triple Negative subtype had a predicted five-year survival of 11.6%
While acknowledging the differences between screen- and symptom detected survival probabilities might be affected by lead time bias, the authors said the model could be a simple tool to help inform newly diagnosed breast cancer patients about their future health outcome.
They noted that existing breast cancer survival models such as the Nottingham and PREDICT prognostic models, had proved useful in decision making for breast conserving surgery and adjuvant therapy, and to assess the impact of different treatments on survival following surgery.
“Assessing the performance of our model using other breast cancer cohorts would be a beneficial and informative next step,” they wrote.
“Given the large percentage of survival variation that was unexplained by this model, gaining a better understanding of what additional factors explain survival outcomes among women diagnosed with breast cancer will require dedicated research studies that include more comprehensive range of factors and/or more nuanced measurements,” they added.