A natural experiment using Scottish clinical data to estimate the real-world effectiveness of adjuvant chemotherapy in breast cancer patients
- Over 4,600 women are diagnosed with breast cancer and around 1,000 people in Scotland die from the disease every year.
- More than 500 patients with early breast cancer are treated with chemotherapy each year in Scotland
- Half of breast cancers are in women over the age of 65, but the average age in trials is closer to 50 and only very fit and healthy women are typically eligible for trials.
It is unknown if chemotherapy benefits seen in clinical trials are the same in typical NHS real-world patients
Reliable methods for estimating the benefits and harms of chemotherapy in NHS patients are urgently required to help patients make difficult decisions about whether or not to undergo chemotherapy.
The SATURNE project aimed to find out if new methods from econometrics and data science could help us address this need.
The very high quality and pre-existing linkage of Scottish healthcare and cancer registration datasets makes Scotland an ideal place to evaluate a new methods for causal inference
70,000 women with an early breast cancer were analysed using data science methods:
- Cox Regression analysis
- Propensity Score Matching
- Instrumental Variable analysis
- Regression Discontinuity Design analysis
Adjustment was made for clinical and molecular stage, comorbidity and socioeconomic status.
- The NHS Predict decision tool is based on a risk model that is valid in Scottish patients.
- It is possible to estimate treatment effect using routine data
- The Regression Discontinuity Design did not perform well in this example
- Patients that are not typically included in clinical trials appear to benefit from chemotherapy to a similar extent to younger fitter patients.
The full results are published in scientific journals
Gray E, Marti J, Brewster DH, Wyatt JC, Piaget-Rossel R, Hall PS and the SATURNE project advisory group. Real-world evidence was feasible for estimating effectiveness of chemotherapy in breast cancer: a cohort study. Journal of clinical epidemiology. 2019 May 1;109:125-32. https://doi.org/10.1016/j.jclinepi.2019.01.006
Gray E, Marti J, Brewster DH, Wyatt JC, Hall PS and the SATURNE project advisory group. Independent validation of the PREDICT breast cancer prognosis prediction tool in 45,789 patients using Scottish Cancer Registry data. Br J Cancer 119, 808–814 (2018) doi:10.1038/s41416-018-0256-x
Gray E, Marti J, Brewster DH, Wyatt JC, Hall PS and the SATURNE project advisory group. Real-world evidence for chemotherapy effectiveness in trial under-represented groups with early breast cancer; a retrospective cohort study. In press with PLOS Medicine
We learned that high quality Scottish healthcare data, combined with leading data science methods make it possible to understand how treatments affect patients in a real-world NHS context.
What does this study add?
Methods for natural experimentation using routinely collected data are embryonic, but new data opportunities are now making it possible to test and further develop new methods that are able to exploit them for patient benefit.
Implications for Practice or Policy
Age and other reasons for for trial non-eligibilty should not be a barrier to patients benefiting from chemotherapy to reduce the risk of breast cancer recurrence after curative treatment.
Current decision tools are based on risk models are accurate as a basis for shared decision making in this context.
Where to next?
We are increasingly adopting new cancer treatments into the NHS before robust evidence with long-term follow-up is available. We believe that methods for causal inference using real world data can complement evidence from clinical trials during the process of early technology adoption.
We have established a Real World Data Analysis Service in an attempt to make such analyses available to ensure that new treatments represent good value for NHS patients.
We would like to see better use of case-mix adjustment when looking at regional variation in practice.
For more information contact Peter Hall at p.s.hall[at]ed.ac.uk