Big data make therapy work better
An international team of researchers has shown for acute myeloid leukaemia (AML) that cancer therapy can be personalised using big data. The authors of the study “believe this paper is a step towards validation of genetic techniques as a route to personalised medicine”.
A retrospective study published in Nature Genetics1 found that if big data had been used to make "statements about an individual’s likely journey through therapy", one in three AML patients would have had their treatment altered and one tenth of all young sufferers could have been spared stem cell transplants. A knowledge bank was developed using genetic and clinical data from 1,540 AML patients who participated in clinical trials in Germany and Austria performed by the German-Austrian AML Study Group. Dr. Peter Campbell, cancer genomics expert at the Wellcome Trust Sanger Institute, and Prof. Dr. Hartmut Döhner, a renowned haematologist from Ulm, were in charge of developing the knowledge bank.
Campell, who is the senior author of the Nature Genetics publication, highlights that the use of big data enables “far more detailed and accurate predictions about the likely future course of a patient with AML than what we can make in the clinic at the moment.”
More expensive but worth it
However, Döhner notes that building and maintaining knowledge banks is costly and time-consuming. “To get accurate treatment predictions you need data from thousands of patients and all tumour types. As genetic testing enters routine clinical practice, there is an opportunity to learn from patients undergoing care in our health systems. Our paper gives the first real evidence that the approach is worthwhile, how it could be used and what the scale needs to be.”
The authors of the restrospective study also discussed health economics issues and compared the number of expensive transplants that would not be needed with the significantly lower costs of genome tests. The fact that the researchers had a few months earlier refined the molecular taxonomy of AML and identified eleven genomic subcategories of AML, might most likely have facilitated the new study.2 The researchers postulated that a knowledge bank approach can be extended to other cancers as well. The researchers' knowledge bank is currently only available to scientists for use in research only. They reanalysed 111 cancer genes (instead of the usual nine AML genes) and the clinical data of the aforementioned 1,540 AML patients. Moritz Gerstung, a bioinformatician from the European Bioinformatics Institute and first author of the Nature Genetics publication, developed multistep statistical models based on the clinical data of the patients, which allowed the probability of remission, relapse and mortality to be predicted.
Next step involves a mega database called “HARMONY“
Lars Bullinger, Heisenberg Professor of Personalised Tumour Therapy at the University of Ulm, is aware of the obstacles that are still preventing precision oncology from becoming standard clinical care. One such obstacle is tumour heterogeneity: no two tumours are alike, even inside one patient. Therefore, the European Innovative Medicines Initiative (IMI) is committed to establishing a mega database called HARMONY (Healthcare Alliance for Resourceful Medicines Offensive against Neoplasms in HematologY) to collect data on seven blood cancers with the goal to improve the care of patients with these diseases. All large study groups store their data in this database, including data from any pharmaceutical companies involved in the respective studies, once they comply with uniform data standards. According to Bullinger, whose role is to supervise this 40-million-euro project, this helps achieve the required statistical relevance for deciphering mutations in tumour genomes and thus enables the rapid detection of promising therapies for certain disease subgroups.
Experiments with the data cloud
The rapidly growing volume of data poses problems for the healthcare sector. Physicians do not always have access to standardised, readable and applicable data. Working with databases such as the European Genome-phenome Archive (EGA) is becoming increasingly difficult because it is almost impossible to download petabytes of data. "This is why we are launching data cloud initiatives at the DKFZ," says Dr. Matthias Schlesner, bioinformatician at the DKFZ. "Nowadays, data provision and analysis cost more than sequencing," also highlighting that the bioinformatic infrastructure of the future will require new technologies on many levels such as compression and new methods such as machine- or deep learning.
The analysis and evaluation of patient genomes has not yet become standard practice at tumour centres, but, as Bullinger points out, the analysis and evaluation of gene mutations that have “a direct therapeutic consequence” is now standard practice. The researchers from Ulm are working specifically on haematology, and use diagnostic panels to study the 100 most commonly mutated leukaemia-related genes. In Bullinger’s experience, some health insurance companies already cover the costs for panel diagnostics while others cover the sequencing of a certain number of genes. Exome sequencing is already on the way to becoming routine application, and, as Schlesner points out, genome sequencing could be heading the same way. Schlesner is the head of the Computational Oncology group at the DKFZ in Heidelberg, where pilot projects related to the clinical application of genome sequencing such as the INFORM register (INdividualized Therapy FOr Relapsed Malignancies in Childhood) are carried out. Such projects are also carried out as part of the Precision Oncology Programme of the National Centre for Tumour Diseases in Heidelberg (NCT POP) and the Heidelberg Centre for Personalised Oncology (DKFZ-HIPO).
A lack of physicians with in-depth knowledge of molecular biology
Genome diagnostics is relatively expensive and takes a long time, which is something that physicians treating acute cases of cancer do not have. Analysing tumour and healthy tissues takes between two and three months, whereas analyses using disease panels only take three days. Moreover, as Lars Bullinger points out, the cost price of diagnostic panels is around 300 euros. The analysis of tissues can possibly be sped up, so that the results could be available in well under four weeks, says Schlesner. If new, targeted medications that cost 10,000 euros per month are used for therapy, Bullinger believes that an initial investment of 1,000 euros/patient for genome diagnostics is an excellent investment if the 50% of the patients who do not benefit from targeted cancer therapies could be identified.
The time required for evaluating solid tumours is usually not a limiting factor, says Prof. Dr. Stefan Fröhling, senior physician and leader of the Division of Molecular and Cellular Oncology at the NCT in Heidelberg and in charge of the NCT's Precision Oncology Programme (NCT POP). In his experience, the major bottleneck is somewhere else: there is a lack of physicians with in-depth knowledge in molecular biology who can assess genome data from a medical point of view.
Just as haematologists often took a pioneering role in oncology, the European HARMONY project, which is taking the public-private partnership initiative IMI into the field of oncology, could also be leading the way. Bullinger sees a huge advantage in the fact that everyone involved in HARMONY collectively covers the entire value chain from academia to health insurance funds, health technology assessment organisations (like the British organisation, NICE), regulatory authorities, patient representatives and pharmaceutical companies.
The patient is in greater focus again
Academic centres will enter data from large register studies or tumour registries into HARMONY. This will broaden the database, because data from patients that have been treated according to standard protocols will also be included. According to Bullinger, this will make it easier to differentiate the results of clinical studies where participants are often not representative of the disease population. This will provide reliable information about the cost of therapy and also about the length of time patients stand to benefit from therapy. Above all, the data will provide information about patients’ quality of life over the longer survival period. Ten to fifteen years ago, these aspects were largely neglected. The researchers hope that the huge data pool will also provide answers for health economics issues. As Bullinger says: “Should I spend the money on financing a very expensive drug that would help a small number of people in a very limited way, or is it better to use the money for useful diagnoses that enable existing drugs to be used more effectively and save more lives?”
New design for clinical trials
Stefan Fröhling disagrees with the representatives of evidence-based medicine who claim that precision oncology is based on correlation rather than causality. He says that laboratory-based evidence of mechanisms of disease-associated mutations has long been provided. Fröhling and Bullinger have identified a different problem: the more progress the molecular characterisation of tumours will make, the smaller the patient groups will become. This makes it difficult to perform clinical trials. Fröhling and Bullinger therefore believe that it is necessary to think about changing the design of clinical trials. Fröhling attaches greater importance to differential therapies rather than the sheer number of patients.