Multiple myeloma is a type of blood cancer that starts in the bone marrow and typically affects older people. Diagnosing myeloma is difficult because early symptoms that people have are also common to people without cancer and can include back pain, bone pain and repeated infections. At present, half of all myeloma patients visit their general practice (GP) doctor more than three times before they are referred on to cancer specialists. These delays are thought to contribute to the poor survival seen with this type of cancer. We want to develop a diagnostic tool that combines patient characteristics, symptoms and blood test results that will alert the doctor to the risk of myeloma so that the appropriate tests can be done. We will do this using the records of patients who have previously been seen in general practice. We will focus on patients over the age of 40 that had at least two full blood count tests within a period of one year. We will then see which patients went on to develop myeloma and those that did not and examine whether there are specific patterns which precede a diagnosis of myeloma.
The objectives of this study are to develop, validate and compare three clinical prediction rules to advance the diagnosis of myeloma based on risk factors, symptoms and laboratory investigations. Patients over the age of 40 with no previous diagnosis of myeloma and two full blood counts within a year will be identified, with the last test being the index date. Patients will be followed for two years after the index date to establish who did or did not develop myeloma. Data will be randomly split into derivation and validation datasets based on practice location. We will use Cox proportional hazards modelling to derive three different prediction rules of increasing complexity. Subsequently all models will be applied in the validation dataset, and discrimination and calibration will be examined. Performance measures will include ROC curves, R2 statistic, D-statistic and calibration plots. Clinical utility of the algorithms will be examined by the use of decision curve analysis. Multiple imputation methods will be employed in order to deal with missing data.
Myeloma diagnosis.
Jason Oke - Chief Investigator - University of Oxford
Constantinos Koshiaris - Corresponding Applicant - University of Oxford
Ann Van Den Bruel - Collaborator - University of Oxford
Brian Nicholson - Collaborator - University of Oxford
Richard Hobbs - Collaborator - University of Oxford
Sarah Lay-Flurrie - Collaborator - University of Oxford