Big Data in Neurosurgery: An Emerging Opportunity
Neurosurgery evolved over decades as technology enhanced our ability to detect disease, image lesions, pharmacologically manage systems, see pathology, probe genetics, and manipulate electrochemical environments. Although many exciting areas of innovation promise to continue pushing the field, data science could be the most broadly transformative.
Data science revolutionized many aspects of the modern world behind the scenes and can do the same for neurosurgery. Data science, also colloquially known as “big data,” is an interdisciplinary field that seeks to ingest, process, and extract knowledge from various data sources in order to generate actionable insights. It encompasses facets of information management such as collection and storage of large amounts of data, pattern prediction via computational analysis and machine learning, cross-platform integration of varied data sources, and analysis in realtime with high-throughput signal detection. Big data has been embraced in non-medical fields for some time, and business has significantly driven developments within the field, transforming processes from supply chain to marketing to knowledge discovery. For instance, we have come to expect that when we search online, we will find what we are looking for within a few clicks, because search engines know our interests, localities, and tendencies. When we buy a product, we expect it will arrive in short order due to streamlined supply, warehousing, and shipping processes. These modern conveniences are possible only with nuanced analytics of incredibly complex knowledge webs.
Data science and medicine ought to be a natural fit—the routine course of clinical care generates volumes of data, and our ability to use those data to better understand disease processes, interactions of systems, and efficacy of therapeutic approaches, is really only limited by imagination and reluctance to embrace discovery. In a recent viewpoint, Escobar and colleagues1 identified six use cases with the clearest opportunities for data to impact healthcare, namely high-cost patients, readmissions, triage, decompensation, adverse events, and treatment optimization in complex or multi-system diseases. Despite modest successes in healthcare and claims remaining largely unproven, there are several areas of promise.
The high prevalence, profound functional, and financial impacts, and robust trial infrastructure for cardiac disease make it a prime area for data innovation. Pediatric cardiology is one interesting example, wherein analytics are applied to make patient-specific recommendations for treatment. In order to improve the quality of care, the Pediatric Cardiac Critical Care Consortium (PC4) collects data on clinical practice and outcomes from each patient’s medical record, analyzes the data, and provides timely performance feedback to clinicians. Data analytics feeds into collaborative learning to foster a culture of continuous improvement. In adults, Shah and colleagues2 tackled a heterogeneous syndrome without known treatment—heart failure with preserved ejection fraction—to develop tailored therapeutic strategies. For each patient, the group collected rich phenotyping data, including 46 clinical, laboratory, ECG, and echocardiographic measures and implemented unbiased machine learning algorithms to cluster patients into groups with more homogenous characteristics, treatment approaches, and outcomes. This study emphasizes how data-driven approaches embracing the complexities of heterogeneous clinical phenotypes can transform treatment and decision-making strategies.
Radiology and pathology are fields ripe for data-driven innovation. At present, the role of big data in radiology relates to decision support to aid radiologists in reading and interpreting images. A recent survey found 89 percent of radiologists said they always use the clinical decision support software computer-aided diagnosis. Pathology has been even further transformed by data. Tumors are now classified much more meaningfully by clusters of molecular markers rather than microscopic analysis. Further, knowing which mutations are carried by a tumor, and thus its clinical responsiveness to different chemotherapies or its radiosensitivity, allow for personalized treatment algorithms.
Neurosurgeons, too, are starting to leverage data techniques to personalize management of neurosurgical disease, prevent complications, and improve outcomes. Several areas are currently under investigation, including development of risk models that combine rich clinical and genetic data and real-time analysis of ICU data to avert deterioration and predict outcome.
High-performance models that combine genetic and clinical data will change the way we practice. The ongoing “omics” revolution opens a myriad of investigative techniques, including whole genome sequencing, single nucleotide polymorphism mapping, and high-throughput proteomic assays, that allow us to probe specifics of both the individual and the disease. To this, rich clinical data, augmented by socioeconomic and environmental data, is added to understand how specific biology interacts to produce outcomes, respond to therapies, and predict complications. Thus, rather than lumping groups of patients together based on broad demographic information or loosely applied criteria from traditional analysis, each individual’s personal risk profile can be considered with great specificity. Such approaches might allow us to more accurately predict who will suffer stroke, develop post-traumatic epilepsy, and recover from infarcts. These insights might help us more intelligently target resources, assign treatments, and counsel patients and families.
Intensive care patients generate continuous data streams, and yet management is typically made based on snapshots of the data without consideration for nuances of waveform and temporal variations. Understanding of symbolic relationships between complex physiological signals and creation of predictive models allows for earlier intervention or prevention of adverse events. For instance, the neurosurgeon, aided by analytic algorithms, may avert impending herniation as a result of early changes in ICP waveform or detect respiratory distress based on changes in ventilator feedback and blood gases. Similarly, acquisition of real-time signals, and integration with other bedside devices, may facilitate closed loop control that will result in earlier correction of problems and ability to more tightly control important physiologic parameters.
Clinical data registries are a tremendous opportunity for innovation, both in terms of how we collect data and how we use data to guide practice. Techniques such as natural language processing and machine learning promise accurate data ingestion minimizing the need for human intervention, and advanced analytic techniques more readily derive insight from large, diverse data sets by considering more broadly the field of potentially contributing variables than traditional regression techniques might allow. These, in combination, open the door for proliferation of registry trials. Although randomized controlled trials (RCT) have long been considered the gold standard for data, for many questions, and in particular for fields that are rapidly evolving (for instance, due to device evolution), RCT are not practical or even desirable. Registry trials are emerging as an evidence standard that allows for more rapid, inexpensive, and high-quality evaluation of clinical questions.
Big data’s promise remains largely unrealized, especially within neurosurgery. We need significant modification in the methods, structures, and institutions of the profession to realize its full potential. Biomedicine—along with other fields—was awakened by major corporations such as Google and Amazon that have revolutionized the Internet roadmap through developing and refining sophisticated data analytics platforms that accurately describe individual human behavior. The reality in biomedicine is there are tremendous stockpiles of high-quality data sitting idle. An abundance of knowledge lies hidden within, and yet only a small fraction has been harvested. The future of biomedicine, including neurosurgery, rests on our collective ability to transform big data into intelligible scientific facts and knowledge.
1 Bates DW, Saria S, Ohno-Machado L, Shah A, Escobar G. Big data in health care: using
analytics to identify and manage high-risk and high-cost patients. Health Aff (Millwood).
2 Shah SJ, Katz DH, Selvaraj S, et al. Phenomapping for novel classification of heart failure
with preserved ejection fraction. Circulation. 2015; 131:269–279. https://doi: 0.1161/