Data mining of cancer vaccine trials: a bird's-eye view

The World Health Organization's Global Burden of Disease statistics identified cancer as the second largest global cause of death, after cardiovascular disease [1]. Cancer is the fastest growing segment of the disease burden; global cancer deaths are projected to increase from 7.1 million in 2002 to 11.5 million in 2030 [2]. Advances in prevention, diagnostics and treatment of cancer have contributed to the improved prognosis for cancer patients: one third of cancers are preventable and another third are curable through early detection and effective therapy [3]. New cancer therapies are subject of vigorous research including the application of new high-throughput biomedical technologies that generate large amounts of data. The related information explosion mandates the use of biomedical bioinformatics.

Researchers and clinicians need rapid access to multiple types of information, including molecular, clinical, and literature databases, and clinical trials registries, as well as suitable data analysis tools. The National Center for Biotechnology Information (NCBI) hosts resources for retrieval and analysis of bio-molecular data. These are accessible through the NCBI's web site [4]. Clinical trials data are available through clinical trials registries. Since 1971, regulatory efforts on clinical trials registration have resulted in a significant level of compliance by both sponsors and conductors of clinical trials [5]. ClinicalTrials.gov [6] has emerged as the largest registry in the world [7]. Currently it contains information on 50,000 clinical trials including 16,000 cancer-related entries. The US National Cancer Institute (NCI) provides access to a clinical trials registry within the Physician Data Query (PDQ) database [8]. The PDQ contains (Dec 2007) 21,000 abstracts of cancer-related clinical trials and regularly exchanges data with the ClinicalTrials.gov registry. Data sharing and direct access to resources (researchers, computers, software, data, research participants, and other) are considered critical for the advancement of cancer research and the improvement of health care. The initiatives such as Cancer Research Network [9] and Cancer Biomedical Informatics Grid [10] provide the framework for integration of various data types and tools for cancer research. Standardized data formats (e.g. demographics, health plan eligibility, tumor registry, inpatient and ambulatory utilization, medication dispensing, laboratory tests, imaging procedures, others) [9] facilitate access to and sharing of data and automated analysis.

Knowledge discovery from databases, also known as data mining, is an emerging field that applies techniques from databases, statistics and artificial intelligence to extract high-level information (knowledge) from a large volume of low-level data. Examples of high-level information derived from low-level data include forms that are more compact (e.g., short reports), more abstract (e.g., descriptive models of the process that generated data), or more useful (e.g., predictive models for estimating values of the future cases) than existing low-level data [11]. Mining clinical trials data usually refers to using statistical and modeling tools for analysis and design of clinical trials. If appropriate clinical trials data (e.g. aims, goals, regimens and conditions, end points, sample sizes, and others) are stored in the registry, data mining can help design better, more efficient trials that require smaller patient cohorts. Standardized data formats, such as XML markup language [12], bring text files into machine-readable form, thus enhancing automated analysis. Both NCI PDQ and ClinicalTrials.gov provide clinical trial registry data in the XML format.

We have developed a data mining approach for rapid summarization and visualization of information from clinical trial registries. This method has been applied to the analysis of cancer vaccine trials and provides convenient means of extraction and presentation of key data about cancer vaccine trials. The significant progress in cancer biology and cancer immunology has not yet been fully translated into successful clinical vaccine applications [13]. Though advances in cancer vaccine development have been reported [e.g. [14, 15]], a wide variety of factors are involved in tumor immune escape, making design and production of effective therapeutic vaccines difficult. Increasing knowledge of possible limiting factors include, among others, dysfunction of the immune system, immunosuppressive effects of tumor microenvironment, production of suppressor T cells, defective antigen processing and presentation, and immunotherapy resistance of established tumors. To utilize the accumulated data and knowledge and translate these into improved clinical trials, integration of basic and clinical immunology and improved data processing capabilities is required. Our data mining approach provides better understanding of the cancer vaccine clinical trials landscape, and enables rapid analysis of the hotspots of cancer vaccine activity, as well as the identification of neglected cancers. This report describes the utility of basic data mining techniques of summarization, tabulation, and visualization applied to the clinical trials repository data.

A large number of records are available in clinical trial repositories but the analysis is limited to simple queries. This data mining application for summarization and visualization of clinical trials data was developed and applied to the analysis of cancer vaccine clinical trials landscape. It enables complex queries of cancer vaccine trials data through data mining. Data mining revealed hidden patterns, trends and biases in the data. The application enables formulating queries and asking specific questions. It revealed the cancer vaccine trials development over time. The most common cancers targeted by vaccine therapy in clinical trials are: melanoma, cervical, prostate, breast, and leukemia. Neglected cancers include bladder, liver, pancreatic, stomach, esophageal, and all of the low-incidence cancers. Our approach provides a bird's-eye view of the cancer vaccine landscape, and enables rapid analysis of the hotspots of cancer vaccine activity, as well as the identification of neglected cancers.

The landscape of clinical vaccine trials has multiple dimensions. These include an historical timeline of cancer vaccine trials, organizations sponsoring vaccine trials, principal technologies, and types of cancers, all of which we analyzed in this study. Many of the fields in the XML files were not included in this study, for example, "minimum age", "maximum age", "inclusion criteria", "exclusion criteria", and others. The selection of fields was defined by the scope of this study: rapid extraction of clinical trials statistics and identification of major activities in the field.

For studies that focus on other questions, such as patient groups, different sets of questions would be asked and different dimensions would be included for selections of relevant fields from the XML files.

The decision by the US Food and Drug Administration (FDA) to publicize information on clinical trials was made to enable public access to data that address questions about the safety of certain drugs [5]. In addition, public access to this data helps counter publication bias, in which positive results are published more easily and faster than negative results [5]. Though a significant quantity of data and information that describe cancer vaccine clinical trials is available, the existing means of converting these data into knowledge are less advanced.

Our data mining process consisted of data preparation and the application of summarization [see [11]] and visualization [see [16]] tools upon extracted specific knowledge from cancer vaccine clinical trials data. By accessing comprehensive clinical trials information using appropriate software tools, several mouse clicks provide access to knowledge that would otherwise require hiring of specialists or consultants. This knowledge is useful for various professionals, including research directors and planners in companies, hospital administrators, researchers and scientists, physicians conducting clinical trials, policy makers, grant officers at funding bodies, and government regulators. By combining public databases of clinical trials, data formatting by XML, and computational analysis and visualization, we have shown that specific knowledge can be extracted, summarized, and presented to the user.

We have implemented a highly flexible system to visualize and mine clinical trial information in the cancer vaccine field. We obtained all data about cancer vaccine trials from ClinicalTrials.gov database. In accordance with the 1997 FDA Modernization Act [17], information from clinical trials on serious or life-threatening diseases must be registered at ClinicalTrials.gov within 21 days of the trial starting date. The ClinicalTrials.gov has become the world's largest clinical trial database and it contains information from clinical trials performed world-wide. Although some cancer vaccine trials may not necessarily be included, ClinicalTrials.gov repository is comprehensive and representative of the world-wide landscape of clinical trials in the cancer vaccine field. Some fields of interest in this study were not included in the original XML files; for example, fields related to specific vaccine technology. We inspected the records, identified vaccine technology, and manually annotated this field using external data sources. In addition, clinical trials records in the back-end database were enriched with fields containing cancer statistics data. Manual annotation of these data has limitations, principally because subsequent studies also need manual annotation. This shortcoming can be addressed either by automation of annotation, or by inclusion of these fields in the clinical trials records. We envision that the power of this system can only increase with the improvement of the content of public databases.