Teaching computers to recognize patterns in data is helping physicians improve the speed and accuracy of diagnosis and further personalize patient care.

To get the right treatment, you first need the correct diagnosis. To help physicians improve diagnostic speed and accuracy, Massachusetts General Hospital pathology experts are essentially teaching computers to mine all the information gathered on each patient and make recommendations about that patient’s care.

Chief of Pathology David Louis, MD, a world leader in the genetic analysis of tumors.
Chief of Pathology David Louis, MD, a world leader in the genetic analysis of tumors

Roughly 70 percent of all hospital decisions about diagnosis and patient care are based on pathology lab tests, estimates Mass General’s Chief of Pathology David Louis, MD. Tests on blood, urine and tissues reveal telltale markers of health and disease. But hidden in the huge haystacks of information generated, he says, “are vast amounts of untapped information.”

For example, “We can now take DNA out of the white blood cells and analyze it for tens of thousands of genes,” says Dr. Louis, a world leader in the genetic analysis of tumors. “Some of the techniques we’re developing allow us to do blood analysis at a level of detail that was unimaginable 10 years ago.”

Advancing Personalized Diagnosis

Mass General pathologists coined the use of the term “computational pathology” to describe their efforts at developing computer models and algorithms—formulas and coding for solving problems—to mine data and make clinical recommendations. Patient data include everything from a patient’s medical and medication history, including exams and test results, plus such external information as genomic databases and related articles from medical journals.

“A diagnosis made three months earlier can make a big difference,” he says. “And also amazing is that it requires no fancy equipment or huge expense.”

Pathologists were central to Mass General being the first U.S hospital to routinely offer molecular diagnostics for cancer, a key to personalizing cancer care by identifying genetic defects. They have made enormous strides in using next-generation genetic sequencing to find more cancer markers. Their current efforts in computational pathology are advancing personalized diagnosis and care for patients with other diseases.

For example, John Higgins, MD, developed a computational model to better use test information on red blood cells to differentiate among the types of anemia. His model allows doctors to diagnose iron deficiency anemia about two to three months earlier than before. This is very important, Dr. Louis explains, because one of the biggest signs of colon cancer is iron deficiency anemia. And, in children, undetected iron deficiency anemia can produce long-term problems in cognitive and neurological development.

Pathology and Data Crunching

“A diagnosis made three months earlier can make a big difference,” he says. “And also amazing is that it requires no fancy equipment or huge expense.”

Drs. Dighe and Baron use the power of computers to analyze patterns found in pathology lab results and other patient information.
Drs. Dighe and Baron use the power of computers to analyze patterns found in pathology lab results and other patient information.

Mass General pathologists Anand Dighe, MD, PhD, and Jason Baron, MD, have been using computers to analyze recurring patterns and trends in patient laboratory test results. “Ten billion lab tests are done in the United States every year,” Dr. Dighe, says, “but they are not always effective because no one is looking at the pieces and tying the loose strings together.”

Drs. Dighe and Baron devised an algorithm that flags laboratory test results with an alert if creatinine, a commonly tested marker of kidney function, is increasing. This is important because trends in creatinine that indicate early acute kidney injury can otherwise be easily overlooked, leading to delayed diagnosis and sub-optimal treatment. They also used data mining techniques to develop an algorithm to discern if a very high glucose test result is due to a potentially life threatening diabetic emergency or whether it is actually due to a common error that can occur in collection of blood. This helps clinicians correctly interpret glucose test results.

Better Forecasting of Disease

They have just received an MGH-MIT Strategic Partnership Grand Challenge grant to expand their work in mining large data sets.

They have just received an MGH-MIT Strategic Partnership Grand Challenge grant to expand their work in mining large data sets. They plan to use more clinical data that will be crunched to enable earlier and more precise diagnosis of other common diseases, such as sepsis, a whole-body response to infection that is hard to diagnose and treat. Mass General and Massachusetts Institute of Technology are jointly committing research funds to support MGH-MIT research collaborations like this one. Dr. Dighe and Baron are co-principal investigators with Peter Szolovitz, PhD, head of the Clinical Decision-Making Group within MIT’s Computer Science and Artificial Intelligence Laboratory, who has decades of experience using these sorts of algorithms.

Meteorologists can predict weather much more accurately these days because they have developed computational ways to analyze immense amounts of current and historical weather data. With further investments in people and equipment at Mass General, Dr. Louis’s vision is that Mass General physicians will similarly be able to better forecast each person’s disease and individualize their treatment.

Please contact us for more information on how you can help further research in pathology that will advance personalized medicine.