Doctors Test Tools to Predict Your Odds of a Disease

ENLARGE

Doctors are testing software to help diagnose and treat patients by ruling out certain diseases. Getty Images

Thomas McGinn, chairman of medicine at a major New York hospital system, is betting he can predict if a patient has strep, pneumonia or other ailments not by ordering traditional lab tests or imaging scans, but by calculating probabilities with a software program.

Dr. McGinn believes using technology to help diagnose and treat patients can reduce the large number of unnecessary tests doctors order and antibiotics they prescribe by ruling out certain diseases. It also could expedite the appropriate care for patients by giving doctors grounds to treat them before lab tests can confirm a diagnosis.

The predictive tool, which pops up on the screen of electronic medical records, prompts the doctor to answer a short series of questions about the patient’s condition. Based on that information, a calculator predicts the probability that the person has the suspected ailment. It may also recommend a course of action.

ENLARGE

Dr. Thomas McGinn, chairman of medicine for the Northwell Health system, at a computer used to access a predictive model for diagnosing patients with pulmonary embolism that he is testing at two of Northwell’s emergency rooms. Photo: Adam Cooper/Northwell Health

An example: Does a coughing patient have pneumonia? The doctor will answer five simple questions, including whether the person has a fever and rapid heart rate and if the doctor can hear a “crackle” in the lungs. A high score based on those responses calls for immediate intervention; a low score, not so much.

Dr. McGinn, who is with Northwell Health, which has 21 hospitals in New York, is running a pilot program to test the system on patients with suspected pulmonary embolism in emergency departments at two of the hospitals he oversees. It is also being tested, with government funding, for strep and pneumonia in nearly three dozen primary-care clinics in Wisconsin and Utah.

As would be expected, many doctors balk at the idea of a computer program telling them how to do their job. The calculator makes diagnosis and treatment decisions seem simple when they really aren’t, says John Beasley, a family doctor for more than 40 years whose Verona, Wis., clinic is participating in one of the trials. He says he ignores the tool when it pops up on his screen.

ENLARGE

Dr. John W Beasley, a family doctor whose Verona, Wis., clinic is participating in one of the trials of the predictive software. He says he ignores the tool when it pops up on his screen. Photo: Stacee L. Goedtel, D.O.

“On one side are the people who want to take a statistical approach to this, versus those of us who say this is a humanistic enterprise and you frequently do stuff that is irrational because it is good for the patient,” Dr. Beasley says.

Some patients come in with multiple medical problems, which could make the calculator less useful or even inappropriate. Someone who is immunosuppressed, for example, may not show a fever even when suffering from pneumonia. That could mean the person is sick even when a predictive program says it isn’t likely.

But proponents of predictive models say doctors are free to ignore calculated predictions, particularly if their instinct is to dig deeper.

For all his work on predictive models, Dr. McGinn says he values the doctor-patient bond. “I love the physical exam and a good [patient] history, and they are more important than any technology,” he says. The ideal is “the marriage of science and instinct.”

Dr. McGinn, who is 55 years old, began his career as a medical resident at a hospital in the Bronx that drew many poor patients. He recalls watching colleagues waste precious resources and make medical decisions he felt were wrong and based more on intuition than research. In the early 1990s, he worked weekends at a Manhattan clinic and was expected to automatically give patients with colds, coughs and flu throat swabs for strep tests and prescriptions for antibiotics. He became fascinated by the notion of creating predictive models for illnesses, which he began working on in his spare time. Since then he has often divided his time between clinical practice and research.

Dr. McGinn led a randomized trial of two predictive models for diagnosing strep and pneumonia at primary-care clinics in New York City. The research found that doctors who used the models wrote about 25% fewer antibiotic prescriptions than a control group of physicians not using the tools, he says. The study was published in 2013 in the journal JAMA Internal Medicine.

Other studies have tested the reliability of the predictive models in correctly diagnosing ailments, he says.

There are now dozens of predictive tools for various ailments. Dr. McGinn and other researchers add new conditions as they complete studies.

In the emergency-department pilot program for pulmonary embolism, a dangerous condition that prompts many doctors to order a CT scan to rule it out, an alert pops up on the doctor’s screen as he is about to order a scan. The alert prompts the doctor to use the calculator to figure out the probability the patient has an embolism.

It is a “useful tool,” says Salvatore Pardo, vice chairman of the emergency department at North Shore University Hospital in Manhasset, N.Y., one of the ERs enrolled in the pilot. Still, he says, doctors are free to override the predictions and frequently do. “There is still a lot of gestalt left” in how his doctors make a treatment decision, Dr. Pardo says.

A new study by Dr. McGinn’s team showed how to use the calculator to predict if a patient will develop C. difficile, a dangerous hospital-acquired infection. Among the questions the calculator asks: Is the patient over 65? Has he or she been admitted to a hospital in the past 60 days. And has the patient been on antibiotics? If the calculator predicts a high probability the patient is at risk, hospitals can monitor the patient more closely, Dr. McGinn says. The study results were disclosed in May at a meeting of the Society of General Internal Medicine.

ENLARGE

Dr. David Feldstein is overseeing the predictive model’s trial for strep and pneumonia in 22 primary-care clinics in Wisconsin. Photo: University of Wisconsin

Sometimes efforts to calculate diagnoses don’t work out. A 2015 study in the journal JAMA Internal Medicine critiqued one tool used to predict whether a person in a hospital is in danger of deep vein thrombosis, a clot in the lower leg which can be deadly. The tool overestimates the likelihood of a clot and lots of hospital patients get high scores. Doctors then treat them as if they were getting a clot and prescribe blood thinners. Dr. McGinn says the tool is flawed for inpatients, but says “you have to know where and when these [predictive models] should be used.

Some resistance to using the predictive model stems from “click fatigue” as doctors deal with a wealth of electronic information, such as best-practice recommendations for treatment, that increasingly pops up on their computer screens, says David Feldstein, an associate professor at the University of Wisconsin School of Medicine and Public Health.

Dr. Feldstein oversees the predictive model’s trial for strep and pneumonia in 22 primary-care clinics in Wisconsin. About half the clinics are using calculators for predicting strep and pneumonia, while the others, the control group, aren't. Since the trial began several months ago, doctors have made use of the predictive model in 20% of cases, he says, but he is trying to increase the rates by educating physicians. “There is a big backlash against clinical decision support,” he says.

It comes down to thinking through what is really best for the patient, says Dr. Beasley, in Verona, Wis. Dr. Beasley, who is a professor of family medicine at the University of Wisconsin School of Medicine and Public Health, acknowledges, for instance, that many doctors overprescribe antibiotics. But sometimes they are needed, even if a calculator says they aren’t.

“I can either prescribe $4 penicillin” on the chance that a patient has a strep infection, Dr. Beasley says. Or he can order a $51 strep test to make certain the person does. For a patient struggling to make ends meet financially, he says he prefers the $4 penicillin.

Write to Lucette Lagnado at [email protected]