October 2018 Edition Vol.11, Issue 10

Data-Sharing Advances the Continuum in Precision Medicine

by Megan Garlapow, PhD

Precision medicine is currently poised to improve treatment outcomes for patients with cancer due to the growing understanding of underlying pathologies. The phrase “precision medicine” has evolved from “personalized medicine”, and the difference is important. The concept of personalized medicine gave rise to an inaccurate notion of treatments tailored specifically to each individual patient.1 Precision medicine, on the other hand, more accurately conveys the aim of tailoring treatments to patients that fit into particular cohorts based on an understanding of characterized variations in pathologies.1

This understanding can arise from several signals or biomarkers, such as DNA or RNA sequence, epigenetic, immunologic, protein, or metabolic signature differences that associate with specific clinical characteristics in a number of patients. This number of patients is a critical detail, the potential size of which has become dramatically enhanced with the proliferation of large-scale data sources, or real-world evidence, facilitated by sharing patients’ electronic health records (EHRs) data.

Sharing EHRs offers the chance to assimilate patterns among patients, providing disease characteristics and tumor signatures for sophisticated analyses. Bobby Green, MD, Senior Vice President of Clinical Oncology at Flatiron Health identified realms where real-world evidence shines: during drug development, post-marketing surveillance, and during patient care. Flatiron Health is also giving attention to external or real-world control arms for clinical trials.2

Even prior to the current era, BRCA1 and BRCA2 genetic variants were used in attempts to predict outcomes in patients with breast cancer, though BRCA1 and BRCA2 genotyping has not always made for obvious best courses of treatment.1 More recently, with advances in genomic sequencing in increasing numbers of patients, multiple genes have been identified with breast cancer treatments that match some of the genetic variants, thus leading to improved objective response rates and other measures of treatment success.3 This tendency stands to be enhanced by development of liquid biopsy approaches that rely upon signals detected within a patient’s blood, rather than by more invasive testing of tissue biopsies.3

An aspect of precision medicine is utilization of analytics such as machine-learning methods to identify patterns predictive of patient outcomes. By combining genome-wide gene expression profiling with in vitro drug sensitivity analysis tests, researchers were able to train machine-learning models to detect molecular markers that were predictive for sensitivity to particular chemotherapy drugs in patients with acute myeloid leukemia.4 Expression levels of certain genes, such as FLT3 and CASP8AP2 and others, appeared to be prognostic for responses to particular inhibitors.4

Studies such as these can help the biomedical community catalog genetic, gene expression, and/or other biomarker cohorts in association with treatments that are effective against tumors showing relevant signatures. To some degree, immortal cancer cell lines can be and have been used for in vitro drug sensitivity testing and biomarker analysis,5 but with data from patients the diversity of information can be enhanced. Predictions can become more refined and strengthened as more data become available, ideally improving outcomes.

A limitation of in vitro cell line drug sensitivity assays is the lack of immunological interaction, which is a characteristic of many in vivo drug-tumor interactions.5 This makes shared real-world evidence all the more potentially useful in developing patient cohorts for precision medicine.

In addition to its applicability to tumor sensitivity, precision medicine potentially allows for reduced risk of treatment-related adverse effects. A meta-analysis of 570 phase II studies containing data from 32,149 patients with varying types of cancer showed significantly better outcomes in patients who were treated based on tumor molecular characteristics than in patients without precision-based treatment.6

Response rates for precision- versus non-precision-based treatments tracked in the meta-analysis were 31% versus 10.5% (P<.001), respectively. Median progression-free survival was more than twice as long with precision-based treatment as without (5.9 vs 2.7 months; P<.001), and overall survival was 13.7 versus 8.9 months, respectively (P<.001). Targeting based on genomic data showed greater survival value than proteomic data did, and overall precision-based treatment also showed a significantly lower rate of treatment-related fatalities (1.5% vs 2.3% for non-precision-based treatment; P<.001).6

Analysis of real-world evidence for precision medicine may be complicated by the challenge of interpretation. Genetic mutations are not always meaningful or obvious in their effects. Laboratory analysis of biochemical signatures may be affected by artifacts or ambiguity. Machine-learning approaches for predictive analytics, also, are diverse in methodologies and, therefore, can be subject to judgment calls if different techniques predict different outcomes or widely disparate rankings of feature importance.3

The structure of many bioinformatics datasets may pose an additional challenge for predictive analytics using some machine-learning methods. For instance, the numbers of samples are typically smaller than the number of samples available for social media analytics.5 More importantly, the set of features is often extremely large, because each nucleotide of a gene sequence or compound in a metabolite profile may represent a singular feature. This is a problem of data dimensionality; the analytic method must identify and evaluate important features from a massive feature set, while often also doing so with a limited number of samples.5 This makes access to data sharing of real-world evidence all the more crucial for healthcare predictive analytics.

A dilemma that can present in precision oncology would be a situation in which clinical trial data and patient cohorts are not developed enough yet to include patients with combinations of relevant biomarker types. An example could be a patient with acute myeloid leukemia whose test results show biomarker classes that suggest appropriate treatments that may not have been evaluated in combination with each other. Additionally, sequential therapies that may be recommended by precision medicine are not well understood.

A growing supply of molecular testing options related to particular treatments has led to the creation of molecular tumor boards at care facilities large enough to support them.7 The boards consist of experts able to weigh a patient’s molecular test results (and other patient information) to evaluate treatments that match the patient’s molecular characteristics and needs. However, at many facilities where patients receive oncology care, in-house molecular tumor boards may not exist. Such facilities may use virtual tumor boards—offsite services that help to advise on appropriate treatment and are often able to deploy efficient technologies and user-friendly interfaces for physicians.7

Data sharing itself is another technological hurdle regarding the use of real-world evidence for healthcare analytics. This is relevant for physicians needing to search for data on cohorts of patients similar to their own patients and for researchers defining patient cohorts based on analyses of biomarkers and outcomes.7 Currently, companies such as Syapse and Flatiron Health offer software platforms to facilitate data storage and sharing of EHRs throughout associated clinics and/or between institutions.

“In order to scale precision medicine, we need data sharing. An oncologist needs to be able to find patients who are similar, clinically and molecularly, to the patient they are treating, and understand the treatments and outcomes for those similar patients,” said Jonathan Hirsch, MS, Founder and President of Syapse. “Sharing data expands the data pool and breaks down the geographic barriers that many health systems face when implementing precision medicine,” he said.

Data-sharing consortia, such as Project GENIE and the Oncology Precision Network, have been developed to test data sharing and its outcomes among facilities.7 Project GENIE has produced initial results, finding that checkpoint inhibition therapies may be useful for certain patients against a broad range of cancers. By 2019, the Oncology Precision Network expects to share data involving more than 100,000 patients per year from more than 200 primarily large US community-based hospitals and hospital systems.7

Patient privacy is a highly critical feature of healthcare data sharing programs and is currently maintained according to federal HIPAA and HITECH laws.7

An issue facing smaller community facilities is the often slower pace of technology adoption, including incorporation of molecular testing into patient diagnostics.7 This is a function of a perception of unclear value to physicians combined with administrative complexity and cost.7 However, one recent analysis of costs for 72 patients showed that molecular testing for precision medicine may be cost-neutral, and testing may increase in accessibility with FDA approvals of molecular tests.7, 8

Dr. Green, however, said he does not see a significant difference between big hospital centers and community cancer centers in terms of the implementation of sharing of real-world evidence. He noted that the challenge of how to develop “a true learning healthcare system where real-world evidence starts to feed back to physicians directly at the point of care” is not specific to types of facilities.

Dr. Green also underscored the general difficulty in obtaining cost data from payers that could be linked to clinical data in real time and suggested the solution will likely involve payers adjusting their systems, which is not a trivial task. One frontier being addressed in the field of data sharing of real-world evidence, according to Dr. Green is how to “validate a real-world endpoint when you’re talking about real-world data.”

Precision medicine is already showing its use in oncology, though some obstacles to implementation remain. However, multiple programs are currently underway to tackle these challenges and improve patient outcomes as safely and effectively as possible.

According to Mr. Hirsch, “last year, the most exciting progress was in the federal government’s clear recognition that precision medicine works, is the future of cancer care, and needs to be expanded and supported.” Mr. Hirsch emphasized recent FDA actions such as approval of the first site-agnostic drug based on a mutation, “and the joint FDA approval and CMS national coverage decision for [next generation sequencing] NGS testing for advanced cancer patients.”

Mr. Hirsch also credited the White House Cancer Moonshot with encouraging data sharing among health systems. “The effects of increased data sharing are still just beginning,” he said, “and will multiply exponentially as more health systems join these efforts.”


  1. National Research Council. Toward precision medicine: building a knowledge network for biomedical research and a new taxonomy of disease. Washington, DC: National Academies Press; 2011. https://www.nap.edu/catalog/13284/toward-precision-medicine-building-a-knowledge-network-for-biomedical-research. Accessed September 22, 2018.
  2. OncoCloud ’18 Presented by Flatiron. 2018. https://www.oncocloudconference.com. Accessed September 24, 2018.
  3. Low S, Zembutsu H, Nakamura Y. Breast cancer: the translation of big genomic data to cancer precision medicine. Cancer Sci. 2018;109:497-506.
  4. Lee S, Celik S, Logsdon BA, et al. A machine learning approach to integrate big data for precision medicine in acute myeloid leukemia. Nat Commun. 2018;9:42. doi:10.1038/s41467-017-02465-5.
  5. Jiang P, Sellers WR, Liu XS. Big data approaches for modeling response and resistance to cancer drugs. Annu Rev Biomed Data Sci. 2018;1:1–27.
  6. Schwaederle M, Zhao M, Lee JJ, et al. Impact of precision medicine in diverse cancers: a meta-analysis of phase II clinical trials. J Clin Oncol. 2015;33:3817-3825.
  7. Madhavan S, Subramaniam S, Brown TD, Chen JL. Art and challenges of precision medicine: interpreting and integrating genomic data into clinical practice. Am Soc Clin Oncol Educ Book. 2018;2018:546-553.
  8. Haslem DS, Van Norman SB, Fulde G. et al. A retrospective analysis of precision medicine outcomes in patients with advanced cancer reveals improved progression-free survival without increased health care costs. J Oncol Pract. 2017;13:e108-e119.

Post a Comment

OBR Archives

To view previous issues of OBR green contact you can visit our archives. The entire library of OBR green articles is searchable.