Last year the Journal of the American Medical Informatics Association ran an article about post-operative nausea, showing that by looking at the right data beforehand, providers can sometimes prevent or ease that misery for their surgical patients. While existing nausea-prediction models only use patient characteristics and medical history, which can’t be changed, the authors identified several points where the choice of anesthetic or pre- and post-op medications had a clear effect on the patient’s level of nausea. They suggested eventually building their model into decision support for anesthesia systems.
The data for the study was laboriously collected by hand—16 data points for each of 2,505 surgical patients at one academic medical center over two years. At that rate, improvements of any kind will come slowly. But with the advent of electronic health records, such studies could be carried on in real time, with treatments tweaked continuously based on a growing volume of data across many providers. In the operating room, experts say the payoff from predictive analytics could be spectacular, since surgery-related expenditures account for almost a third of health care spending.
“I think there’s a tremendous potential in surgical analytics,” says Andrew Watson, M.D., medical director of the Center for Connected Medicine at the University of Pittsburgh Medical Center, head of the informatics committee for the American College of Surgeons and a colorectal surgeon. “Procedures are expensive and invasive and carry risk, so whatever we can do to understand the patient—genomic and proteomic data, history and long-term outcomes—is critical.” Such data could be used to show whether the procedure is right for the patient, whether the right surgeon is performing it, what types of interventions will most benefit the patient before, during and after the operation, and what to expect during the recovery period and afterward.
“The Holy Grail for most quality improvement collaboratives is to not just understand population-level risks, but to drill down to the individual patient level and tailor your therapies,” says Michael Gaies, M.D., a pediatric cardiac intensivist at the University of Michigan and head of the Pediatric Cardiac Critical Care Consortium (see below).
To find that grail, providers must produce better data than they do now, pool it together and analyze it in new ways. It’s impossible to bake a cake when the ingredients stay in their original containers, and impossible to do predictive analytics when information stays siloed. Stanford University anesthesiologist Loren Riskin, M.D., wants to break down those siloes. “We have lots of isolated outcomes registries that we can pull data from—which providers have the highest and lowest rates of nausea, which surgeons have the highest and lowest rates of wound infections—but there’s not enough data to do large research projects,” she says. “Some patients do well and some patients really struggle, and we want to be able to predict who we’re really going to help.”
Riskin’s husband Dan, a critical care surgeon, heads an analytics company called Health Fidelity, Palo Alto, Calif., that is working on ways to use the full clinical record, including free-text notes, for sophisticated analytics. While the company’s initial focus is on appendectomies and hernia repairs, common procedures with plentiful data, Riskin’s eye is on patients in rarer subgroups where only “big data” can make a real difference. “If you look at a patient with diverticulitis and five other medical problems, no one knows what’s the best treatment because no one has had enough data to study it,” he says. “These are areas that are ripe for intervention.”
Though the average hospital may not have in-house analytic capacities, there are at least two key steps all providers can take to put themselves in a position to benefit from the efforts of others.
The first is to be as consistent as possible in data collection, says Vitaly Herasevich, who studies intensive care informatics at the Mayo Clinic. While studying predictors for sepsis, he discovered that many patient records were lacking certain vital sign data and lab results that were only collected when sepsis was suspected. As a result, patients who didn’t develop sepsis were under-represented. “All algorithms should be tuned to the environment and the available data,” he says.
The second is to make sure that agreements with clinical system vendors include access to the raw data. “Then you’ll be able to combine it with other data sources and generate reports that might be helpful,” Herasevich says.
Though predictive analytics for surgery are in their infancy, they’re already showing promise, as the following examples show.
Life-saving in real time
Babies and children with heart problems are among the sickest of patients, and even small changes in the way they’re cared for can have big effects. But compared with the adult population of cardiac patients, there aren’t enough of them in any single hospital to compare cases and draw meaningful conclusions—especially not in real time when changes in care would make a difference.
The Pediatric Cardiac Critical Care Consortium (PC4) is harnessing the power of electronic information and analytics to compare cases across hospitals and someday to adjust care protocols in real time. Unveiled at a medical meeting last year, it started collecting data from participants in August. Data collection software is provided by CardioAccess, Fort Lauderdale, Fla., which enables providers to submit electronic information to several other existing registries and can pull information from those registries into PC4. Analytics are performed by ArborMetrix, Ann Arbor, Mich., which adjusts the data so that each center gets an “apples to apples” picture of how it compares with other centers.
“We’re light-years behind adult specialties, but we now have at least the infrastructure to begin the task of giving clinicians [actionable] information in the relatively near future,” says Michael Gaies, M.D., the consortium’s director,
Early participants include All Children’s Hospital (St. Petersburg, Fla.), Children’s Hospital of Alabama, Children’s Hospital of Atlanta, Children’s Hospital of Philadelphia, Children’s Hospital of Wisconsin, Cincinnati Children’s Hospital and Medical Center, Lurie Children’s Hospital of Chicago, Medical University of South Carolina Children’s Hospital (Charleston), Texas Children’s Hospital (Houston), Toronto Hospital for Sick Children, University of California San Francisco Benioff Children’s Hospital and University of Michigan C.S. Mott Children’s Hospital (Ann Arbor), where Gaies serves as an assistant professor and pediatric cardiac intensivist.
Several registries, including one from the society for thoracic surgery and the American College of Cardiology, collect some data on pediatric heart patients. “All of them have been great and they provide very useful information, but not in real time,” says Gaies. “Real-time access to data is important for people who want to make rapid improvement.” PC4 collects data not only on the surgical procedures themselves, but on interventions and therapies that patients receive afterward, as well as on complications.
Between 500 and 600 cases were submitted to PC4 during the first two months of data entry, from five hospitals. By year-end, Gaies expects to have 40 to 60 new cases per month from 15 hospitals, and he’s actively recruiting more participants. His goal is to have 40 to 50 centers contributing data by the end of 2016, which he says would cover a large proportion of pediatric cardiac surgery programs and most of the relevant cases.
Just one example of how hospitals could use PC4’s data: to assess their use of ventilators. “Nearly all our patients who have surgeries come back on a breathing machine,” says Gaies, and those ventilator times range from one to three days depending on the complexity of the case. Higher vent times can lead to higher complication rates and longer stays. He believes his hospital’s vent times are longer than other institutions’, on average, but he doesn’t know for sure. “We have some ideas how to shorten the time, but we shouldn’t invest resources to change it until we know it’s a problem,” he says. “As soon as we start to get accurate, risk-adjusted data from other hospitals, we can approach it with a multi-pronged attack. Without that data, we would be flying blind.”
Gaies estimates that about 30 percent of the registry information is extracted directly from the electronic health record, and the other 70 percent is entered manually by clinical research nurses or other helpers. “The ideal would be to have as much data as possible extracted directly from the EHR, assuming it could be done accurately,” Gaies says. New cases go into a depository, which loads them into the registry every day at 4 a.m. Aggregated data is updated daily so participants always have the most recent comparative data.
Pediatric heart patients differ from adults in their time horizon, and PC4 plans to address that issue. “We want to look at mortality, but also longer-term outcomes like the reduction of complications that could lead to cognitive and developmental problems,” Gaies says.
Shoulder dystocia (SD)—where the baby’s shoulder gets stuck during delivery—is one of the most dreaded complications of childbirth due to its unpredictability. Though things usually turn out fine, in a few cases SD can lead to significant injury for both mother and child. Injuries to the baby, called brachial plexus injuries, can cause long-term disabilities if they’re severe and not properly treated, and despite their rarity they’re a major source of obstetric malpractice claims. The hazards can be averted with a scheduled caesarean section, but since SD arises in only one to two percent of deliveries and permanent brachial plexus injury occurs in one delivery in 5,000, or fewer, the C-sections themselves would cause more problems than they prevent. More than 90 percent of patients have at least one risk factor for a delivery with SD, as defined by the American College of Obstetrics and Gynecology, but that information alone is inadequate for predicting the occurrence of either SD or injury.
If cases with an elevated risk of brachial plexus injury could be accurately identified beforehand, the risk-benefit equation of a C-section could change substantially, and that’s the objective of a Web-based screening tool from fetal monitoring company PeriGen, Princeton, N.J. Its PeriCALM Shoulder Screen is Web-based and enables physicians to assess patients for seven known risk factors such as maternal weight and height, fetal weight, the presence of maternal diabetes, and whether the mother has had SD in prior deliveries. A study of about 20,000 patients, published in 2012 in the American Journal of Obstetrics and Gynecology, showed a drop of more than 50 percent in the rate of SD, with no increase in the rate of primary C-sections, in the group evaluated using the tool, compared with a control group that was not screened.
“It takes a tremendous amount of research to identify whether there are commonalities you can identify ahead of time to predict certain outcomes,” says Kathryn Townsend, PeriGen’s director of client insurance programs. The tool looks at how the factors interact to increase the risk of brachial plexus injuries if SD occurs. It classifies patients into low, intermediate and high-risk categories based on the experiences of previous patients with similar scores. The high-risk group can be counseled on the relative risks of a scheduled C-section. Users of the tool can also enter information on what procedures and treatments were used, and on the outcome of the delivery.
St. Vincent’s Medical Center, Bridgeport, Conn., was encountering one or two cases of birth trauma a year out of 1,200 deliveries, and the majority of the injuries stemmed from SD complications. In 2008, the hospital’s obstetrics staff started using the SD screening tool as part of an organized effort to reduce birth traumas. Patients are evaluated initially at the 12th week of pregnancy, and then again at 36 weeks. Those with a positive score on the screening tool are further evaluated using ultrasound. Since adopting the tool, the hospital has averaged two to three percent of deliveries with SD, but not a single injury arising from them. While the occurrence of SD itself still can’t be predicted, the screening tool has helped identify women whose babies are at particular risk for SD-related complications, and gives them the opportunity to consider a C-section.
“The value of the screen is to consolidate all the known risk factors for shoulder dystocia,” says William Cusick, M.D., chair of obstetrics and gynecology. “It helps the clinician and the patient become better informed about factors that increase risk.”
Improving cardiology guidelines
Best-practice guidelines, whether from professional societies, national quality organizations or federal research agencies, are based on the accumulated experience of specialists who have seen hundreds or even thousands of patients through their procedures and treatments. They are a vital step for improving quality. But cardiologists at the University of Pittsburgh Medical Center are taking them one step further, by starting to gather long-term follow-up information.
“The lack of bad outcomes doesn’t mean all patients are receiving the right procedure at the right time,” says Oscar Marroquin, M.D., director of UPMC’s Center for Interventional Cardiology Research. “We want to know how they’re faring over time, and whether the procedure resulted in the intended outcome. All of that information takes time to accrue.”
UPMC cardiologists are doing regular follow-up with 5,000 patients who have had interventional cardiology procedures. Marroquin expects that they’ll have enough information within the next year or two to begin to assess, and perhaps recommend changes to, the clinical guidelines they use from the American Heart Association and the American College of Cardiology. He says it’s particularly difficult to identify any long-term harm from a procedure without having a large number of cases to analyze, and up until now, the data just hasn’t been available despite registry activities conducted by the ACC, the Society of Thoracic Surgeons and other organizations. By collecting long-term data routinely, using inpatient and outpatient EHR systems, UPMC is building the robust database necessary to draw valid conclusions.
The data points themselves are fairly simple: vital signs, whether patients have been hospitalized or required more testing for their heart conditions, which medications they are taking and how well they’re adhering to the regimen, and a quality-of-life questionnaire that the patient fills out periodically. The collection process involved the I.T. department in some tweaking of the EHR system, which was not designed to serve as a data collection tool for longitudinal care. “We’ve created new fields that specifically address the handful of endpoints that we’re interested in,” Marroquin says.
UPMC has started to apply the same methodology to collecting data on cardiac surgeries and is now following about 1,000 patients.
ACS Risk Calculator: Harnessing big data
A new Web-based surgical risk calculator from the American College of Surgeons—free to all comers—lets surgeons and their patients assess risks for almost any procedure that has a CPT code. Based on about a dozen details—demographics and medical history, mostly yes-or-no questions—the calculator shows the relative risks of death, various types of complications and return to surgery, plus a projected length of stay.
The risk calculator made its debut this summer, and within the first couple of months it attracted thousands of visitors, says Clifford Ko, M.D., director of the ACS’s National Surgical Quality Improvement Project, on which the calculator is based. It uses 10 years of data from 500 hospitals, representing millions of patients.
Each hospital contributes data to the ACS registry using ACS-trained, on-site registrars who compile all the data so that it’s clean and consistent. The Cleveland Clinic alone has five ACS-trained personnel who only work on related NSQIP data submission.
Ko, a colorectal surgeon, piloted the calculator with his own patients. “It allows me to have an individualized discussion with Mrs. Smith, who needs a colon resection, is 87, and has a congestive heart failure, diabetes and a body mass index of 47,” he says. “Sometimes the calculator shows that the risks are higher than I thought, and I can start to mobilize the cardiologist or the ICU before the procedure. It also lets the patient decide whether she wants a procedure. Maybe she’s 101 and the risk of mortality is 90 percent. That calculation can go into our shared decision-making.”
The risk calculator lets all hospitals and physicians benefit from the data that’s been helping NSQIP’s participating providers improve their surgical quality. Ko says 83 percent of them have improved their outcomes, including mortality rates and infection rates, by benchmarking against their fellow participants and pinpointing problem areas. Both the NSQIP activities and the risk calculator will continue to be refined as more data is added.
NSQIP started in veterans’ hospitals in the 1980s, when there was concern about the VA’s quality of care and pressure to send its patients to private hospitals.
Register or login for access to this item and much more
All Health Data Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access