Machine learning cuts surgical infections by 74 percent

By applying predictive analytics and machine learning techniques to patient data and real-time data from operating theatres the University of Iowa Hospital in the US managed to reduce the incidence of wound infections acquired during surgery by 74 percent.

The hospital has now spun-off a company, Dash Analytics, to commercialise its technology.

Dr John Cromwell, associate chief medical officer at University of Iowa Hospital, and now also CTO at Dash Analytics, initiated the project in 2012.

Speaking to Computerworld at Tibco Now in Las Vegas, he said: “We started work with the hypothesis that if we could predict which patients would get surgical infections we could change the wound management strategies at the time of surgery to reduce the risk of infection.

“Surgical infection in the US is the number one hospital infection and carries the most morbidity,” he said. “It is also the most expensive type of hospital infection to treat.”

(The most recent, 2011, statistics available from the US Center for Disease Control put the annual number of surgical site infections at 157,000 out of a total of 722,000 hospital-acquired infections).

Cromwell explained that a number of factors contributed to the risk of surgical infection: “The patient might be malnourished, or morbidly obese, or they might be on medications that supress the immune system. The duration of the operation influences the risk, and whether you keep the patient warm throughout the entire operation makes a difference.

“We designed a real time tool that uses the medical record data plus the real time data from the operating room to provide some decision support to the surgeon at the time of the operations to change the wound management strategy.”

He said surgeons had basically two options at the time of surgery that could be applied to mitigate the risk of infection: Leave the wound open or use a technique known as negative pressure wound therapy where the wound is sealed and a vacuum applied to it.

“Negative pressure wound therapy can reduce infections significantly if it is applied to the right patients, but it is not inexpensive so we wanted to be selective. There are different hypotheses as to why it works.

“We used the analytic tools to determine which patients should get negative pressure therapy and within two years we had reduced the surgical infections by 58 per cent and in three years by 74 per cent.


“We were very surprised. The result was far better than antibiotics. … And we probably saved the hospital between [US]$1.2 million and $2 million per year.”

At the start of the project Cromwell said he was writing code himself but quickly realised that enterprise grade software was needed and opted for Statistica, which is now a Tibco product after going through several ownership changes. It provides data analysis, data management, statistics, data mining, machine learning, text analytics and data visualization procedures.

Built with Statistica

“We have built all our models on Statistica,” Cromwell said. The enterprise version is an end-to-end tool where you can do very quick explorations of the data and if you find you are going in a direction that makes sense you can build out that entire deployment on the same system. It has worked well for us.”

After speaking about the system at various conferences, and being reported in the Wall Street Journal, Cromwell said the hospital starting getting enquiries from other hospitals but did not have the resources to assist them so decided to create Dash Analytics to commercialize the technology.

“It is at a very early stage,” he said. “We are working with Tibco on deployments in other hospitals. We have a bunch of target hospitals where we are doing implementations.”

He added: “We are in talks with several very large companies that have an interest in this so I don’t know for now long Dash will be a very small company.”

Dash Analytics announced in April that its technology had been chosen for a CDC Epicenters Program research trial to assess, across multiple hospitals, the efficacy of using negative pressure wound therapy in preventing surgical site infections. Cromwell said five CDC Epicenter hospitals were participating.

Other applications

He said Dash Analytics was also investigating how machine learning could be applied to other medical procedures: Blood transfusion, clostridium infections, and delirium.

It is hoping to identify patients coming in for elective or semi-elective surgery that are most likely to need a blood transfusion and enabling treatment to be provided ahead of surgery to reduce the likelihood of them needing blood. “It might be as many a quarter those of those needing a transfusion can avoid having one. We think we can have a big impact,” Cromwell said.

“Clostridium infection is a big problem in the US and can be lethal. It is activated through the use of antibiotics that knock out the normal flora in the gut allowing clostridium to grow in its place. And it is very resistant to antibiotics. We are trying to identify patients at risk using machine learning.”

Delirium, he said, was “a form of brain failure and has become much more common as the patient population has aged. Patients who develop it have a mortality rate of up to 40 percent over succeeding months.”

Original article by Stuart Corner
This post was curated with edits by Gordon Fletcher, Principal Consultant(Engineering & Mobile Technology)  at Compumagick Associates can be reached at, @compumagick