In histopathology we are always trying to improve ways of predicting prognosis in tumours, especially tumours like breast cancer where overall stage is similar amongst many tumours (i.e. node negative breast cancer). Simple methods on H&E sections have been very effective, the Nottingham prognostic index in breast cancer being an excellent example, but the advent of molecular pathology has led to the development of other methodologies. A commercially-available test for prognostication in breast cancer is Oncotype DX which analyses the activity of 21 genes in tumour tissue. Oncotype DX is licensed for use on patients with node-negative, ER-positive HER2-unamplified breast cancer and there are well-validated studies of its prognostic power. The problem, in both private and NHS healthcare systems, is that the test is relatively costly (about £2500 in the UK) and although NICE originally sanctioned its use the NHS declined to fund it (it may now be available on a loss leader basis from Genomic Health, the parent company).
An interesting approach has been taken by the Department of Pathology at the University of Pittsburgh where they have ‘reverse engineered’ the Oncotype DX score using the readily available parameters of Nottingham prognostic index, ER, PR, HER2 and Ki67 staining. They sent over 800 samples for Oncotype DX and then used linear regression to derive equations that gave a surrogate ‘Oncotype DX score’ from the available parameters. They validated these equations on a further 200 cases and have made the equations available on a webpage. Interestingly they didn’t just use their equations to replace Oncotype DX but used them to screen out ‘obvious’ low and high recurrence risk cases and then sent the cases in the middle for Oncotype testing.
The problem with all of this type of prognostication is validation with large datasets. Although 800 cases to derive the equations and 200 test cases sounds a lot it really isn’t sufficiently large to prove entirely robust validation of the Pittsburgh equations. The correlation between the results from the Pittsburgh Magee equations and the real Oncotype score is statistically good but it isn’t perfect. Even the Oncotype DX validation studies do not have huge numbers with a total of 3707 patients in 5 separate studies (and one of those studies had a case control methodology) . I think the real value of these tests will only become apparent when clinical evidence has accumulated over a few more years.