The Promise of Big Data for the Medical Specialist
The United States’ healthcare expenditure accounts for roughly 19 percent of the gross domestic product, with more than $1 trillion spent by the Center for Medicare Services alone. Every healthcare interaction in the modern world generates data of some sort: a transaction, textual data, telemetry and outcome metric. The petabytes of non-imaging data are a treasure trove of information as are the increasing amounts of imaging and video data. The data amassed is far more than humans can analyze without assistance.
Data analytics have allowed companies to understand the marketplace in ways that could not have been predicted, such as understanding habits of customers and clients en masse. These analytic tools can be applied in medicine as well.
Traditionally, medicine moved forward by observation of effects or proposed mechanisms that generate a hypothesis, followed by testing and validation. This present ocean of data defies this customary model, because of the innumerable variables and interactions involved.
The data science tools that have proven beneficial to business and engineering science hold promise for medicine. The prize?
- Saving even a fraction of a percent of healthcare expenditures will be beneficial to payors and society.
- Detecting previously unnoticed levers can yield improvements in treatment efficacy and quality of life.
- Identifying harmful interactions can yield reductions in morbidity. Inherent to this approach is the presence of a feedback loop that allows for newer and more relevant data to be fed back to the system.
In the neurosciences, many horizons beckon. With the explosion of interventional stroke management, one of the key opportunities is improving logistics of care delivery – getting patients to intervention in an expeditious manner for the delivery of effective care. Data science and data transfer can potentially hasten the triage process, reducing time to treatment and giving back to patients.
Approaches to understand the electrical signaling of the brain are also on the rise – modeling the fundamental neuronal circuitry of the brain will benefit from machine learning to assess interconnectivity – allowing functional neuroscience to continue its growth in theory and treatment.
Impacts in neuro-oncology are promising as well. Algorithms appear to be able to discern genetic differences between gliomas by relatively subtle MR features – an elusive feat. Careful analysis of neuro-oncologic Big Data may also allow clinicians to better tailor therapies to specific tumor and patient subtypes to improve outcomes, or so-called personalized medicine.
One may be tempted to believe that these spectacular successes sound the bell for a paradigm shift in medicine, but the challenges restraining the widespread adoption of machine learning in medicine are no less impressive.
Much of the data in healthcare is not organized in a fashion that yields itself to analysis. Others establish provenance of patient care, rather than careful analysis of patient conditions. Much of currently recorded data is summative, rather than granular. We limit our assessment/interpretation of patients to observations we believe are important, rather than relying on raw data that may hold salient information that we do not, or are incapable of, perceiving.
The incentives for developing intelligent care are also unclear. The payors of the healthcare system will clearly benefit from efficiency gains, but providers – doctors, nurses and other allied health professionals – do not have as clear a path to benefit. Developing insights to disease – producing data-driven “classifiers” that can detect differences in patients and improve care – is not simple.
Merely “boiling the oceans” to find patterns may be an attractive approach to investors, but clinically meaningful algorithms can be elusive. As an example, one might find a signal that discerns the age of a patient from a hand radiograph – information that is unlikely to impact in costs or care delivery.
Large amounts of high-quality data are required to feed data scientists and algorithms to devise reliable, reproducible outcome models; for many diseases, the known samples are relatively small and high-quality datasets are usually even smaller. Many of the clinically interesting questions revolve around smaller populations; the healthcare informatics systems are not designed to capture and concentrate such data.
For these promises to be realized, data collection methods need to be improved; what is required are high-quality input methods that aren’t duplicative or disruptive. Data scientists need clinical insight to understand what questions are impactful now – where effort will yield improvement in understanding, care and cost. Clinicians need time and resources to work with data scientists. Allying the analytic speed and thoroughness of machine intelligence to the insight and intuition of human intelligence is likely the most promising approach for the near future.
We need a new generation of physician-data scientists to realize the medical promise of Big Data. Our patients deserve it.
No upcoming events
Be the first to reply using the above form.