Big Data Versus Doctors
Annual Cost To The Usa
- 15% of patients experience diagnostic errors
- 1 in 30 cancers are misdiagnosed
- 1 in 5 cancers are misclassified
- 10% of patients deaths are caused by diagnostic errors
- cancer diagnostic error rates may be increasing
- 44% of radiology malpractice claims for diagnostic errors are cancer-related
Interfaces for healthcare IT remain grossly outdated. According to research, most all leading Electronic Health Record (EHR) interfaces earn an “F” rating from doctors. Legacy PACS interfaces used by Radiologists allow mostly viewing of only 2D slice images with limited 3D modeling, and provide physicians very limited interactivity with patient data.
Doctors today expend vast hours to simply gather the data they need to solve or address a clinical problem. Due to interoperability limitations, the data is often not available or must be faxed between health systems. Legacy interface chaos also forces doctors to surf through datasets in separate data silos, each with separate interfaces and logins.
OUTDATED TEXT-BASED REPORTS
Radiology reports and other types of medical reports remain in legacy text formats, often stored as separate PDF files in legacy systems. Legacy text reports often generate confusion and unwarranted anxiety for patients. Worse, legacy text formats are a poor form of physician communication and can be a source of diagnostic errors.
SUBJECTIVE DIAGNOSTIC "IMPRESSIONS"
Diagnoses today are still primarily made only by subjective “impressions” based on a doctor’s wealth of expertise and experience. No tool has yet been significantly adopted in clinical practice to allow doctors, such as Radiologists, to quantify their diagnoses using modern algorithmic tools.
In recent years, many groups have attempted to use “artificial intelligence” algorithms (aka, neural networks) on single medical images to automate “black box” diagnostic decisions made by Radiologists today, coined “AI Automation.” Although it has proven powerful in closed-system training, these brittle algorithms perform poorly in the open and less predictable environments of clinical medicine. Further, trained algorithms show poor transferability from one healthcare system to another, and are not interpretable by medical experts. In a recent ACR survey, 93% of radiologists say these tools are unreliable in open-system clinical medicine. Top experts believe it will take >10 years to safely implement AI Automation into healthcare, leaving it in the “trough of disillusionment” on the Gartner Hype Cycle.
Yet, “Radiomics” research also supports evidence that patterns of big data from medical images, sometimes combined with other data such as genetics, can provide biometric predictions of patient outcomes, such as the likelihood of response to a certain drug treatment. However, these methods use outdated standard segmentations that have been shown repeatedly to lack even the most basic measure reproducibility. Radiomics and Radiogenomics have thus seen limited successful translation into clinical medicine.
We know that powerful new big data technology provides a great opportunity to solve massive problems. Failure is simply not an option.
Everyone deserves better.
And especially patients.
We Can Solve It. Together.