Targeting microbiology lab efficiency with AI

Amy Carpenter Aquino

August 2020—Bringing an automated culture plate reading system into the Hennepin County Medical Center microbiology laboratory was never a question of if but when.

“We need artificial intelligence to help us with active decision-making processes in the lab,” says Glen Hansen, PhD, medical director of clinical microbiology and molecular diagnostics at Hennepin County Medical Center in Minneapolis. “The difference between robotics and AI is that we are now at a point where we want these instruments to help us with decision-making processes, not just robotic functions.”

His laboratory’s interest in automated plate reading is born of the trio of aims that labs everywhere are trying to achieve with automation, he says: “How can we get more testing out of the current staffing levels in our laboratory, how can we decrease turnaround times, and how can we introduce tools into the laboratory that allow the microbiology work to be done more efficiently, with fewer errors, and more time to protect the skill set of my laboratory techs for the areas we want them to look at?”

When Dr. Hansen’s team acquired the Automated Plate Assessment System (APAS) Independence (Clever Culture Systems) to process urine culture readings in December 2018, one of the selling points was the modular option that could be adapted easily to his laboratory’s existing layout, he says. Another was the pipeline of modules for methicillin-resistant Staphylococcus aureus (under FDA consideration), vancomycin-resistant enterococci, and zone size reading of antibiotic sensitivity discs. A study performed at Johns Hopkins Hospital found that the APAS demonstrated high accuracy compared with manual readings of MRSA cultures and detected a low level of positives missed by manual reading (Gammel N, et al. Poster presentation at ASM Microbe Online 2020).

This year, Hennepin finalized its purchase of the APAS Independence, which uses AI to interrogate colonies for size, pigment, and granularity. “We believe systems like this are needed in the lab,” Dr. Hansen says.

The APAS Independence, which the FDA cleared in 2019, screens urine cultures for significant growth and removes negative cultures from the workflow. Dr. Hansen and colleagues in 2019 reported the results of the first U.S. use of the instrument to screen urine cultures. In the study, APAS categorized cases as significant growth, marked for review, no growth, or errors. Of 720 urine cultures evaluated, APAS identified 98 percent (370/374) of cases reported as no growth by the clinical lab. APAS identified as errors (caused by defects in the media) four cases that the lab reported as no growth. Thirty percent of the 720 cultures were marked for review by the lab. The combination of significant growth plus cases marked for review by APAS correctly identified 100 percent of all positive cultures (Hansen G, et al. Poster presentation at ASM Microbe 2019).

“We can target the work better with these tools,” and instruments make it possible for technologists to apply their skills where they count, “whether that’s decreasing turnaround times or focusing on cases that are growing,” Dr. Hansen says.
“I don’t want to spend 50 percent, let alone 80 percent, of talented tech time to read a negative plate. Lab automation allows us to focus the work of our techs on cases that matter.”

A shrinking workforce means maximizing resources, he says. “I’ve got eight talented people in the lab who work my day shift, whom I depend on to come to work Monday through Sunday and work from anywhere between five in the morning to five o’clock at night. How can I use their skills more effectively, so that if we’re asking them to do four things in a day, and I can provide them with more tools, they could do seven things in a day?”

The APAS Independence has had an impact, Dr. Hansen says. “The ability for us to put urines through the APAS has changed our entire workflow. Because of the way my benches are organized in the lab, the techs sit at a bench and see all the cases that come in. If I can remove the urine workflow from those benches, my benches become a lot more efficient, and as a result there are cases we can get to earlier.”

For labs with dedicated urine benches, he says, an automated culture reading system could provide enough efficiency to allow for a 50 percent or greater reduction in the number of urine benches.

“With the pandemic, laboratories are under more financial constraints than they’ve ever been before, and staffing levels are going to be assessed like they’ve never been before,” Dr. Hansen says.

In Hennepin’s microbiology laboratory, the primary job of some staff is to look out for urine cultures that could be loaded onto the APAS after incubation. “Right now, the system does not do incubation,” Dr. Hansen says. “We do incubation in our normal incubators, and then we pull the plates over in the morning. The plates are time-stamped so the techs know how long they’ve been incubating, and then we put them into the system.”

APAS sorts the plates and autoclears negative plates from the workflow within 16 seconds. FDA clearance is for all urine cultures with detected positive growth to be sent back to the microbiologists for review. “So it doesn’t autocall systems,” Dr. Hansen says. “It identifies the growth patterns that are likely to be of clinical significance.”

He and his team have worked with Clever Culture to code the system to allow the technologists to understand what is a high-probability-positive urine culture versus a category they call “doubtful.” “So we have used the system to fine-tune it to provide a little more diagnostic decision-making in the lab, but those [positive] plates right now come back to the workflow benches for the techs to review.”

More than half of the urine culture workflow were negatives or was considered to be of limited clinical significance.

“Preliminary experience so far”—preliminary because the APAS isn’t yet connected to the reporting system—“suggests efficacy increases and a reduction in hands-on times with urine culture reporting of around 15 to 20 percent,” Dr. Hansen says. “For us, that equates to between a one- to two-hour savings per day in urine culture handling.”

“At the end of the day, it allows us to get through the workload more effectively and quicker than we’d been able to do without it. That in and of itself has endeared the system to the work staff.”

Time saved from the automation of urine culture screening allows other culture screenings to be completed more quickly. “If I’ve got a tech sitting at a bench that has urine, throat, and respiratory cultures, there are a series of cases that, without the APAS, we might not get to until after lunch because of how the benches are configured,” Dr. Hansen says, citing as an example a respiratory sample from an ICU patient. “Because we can get to that case before lunchtime, if that case represents a significant case where the tech would have to do an antibiotic susceptibility on it, all of the work that flows from that case is now also adjusted. We believe that the susceptibility parameters of doing conventional culture in the lab will also shift because of the APAS.”

It didn’t take long to see the benefits, he says, “and the pandemic has only strengthened our understanding of how tools are needed in microbiology to get through the day.” The Hennepin microbiology laboratory, like other labs in the U.S., has had to shift staff to the molecular laboratory to help with the COVID-19 testing workload. “Tools like the APAS have unquestioningly allowed us to get through the day more effectively with lower burnout, lower stress, and, quite frankly, higher accuracy as you start to work people longer hours.”

“This is a system that allows us to target their talents,” he says of the staff.

APAS’ high level of accuracy in ruling out cases of limited significance is not what appealed most to Dr. Hansen. “I expect it to work. I’m not going to bring anything into my lab that doesn’t work,” he says. “What I want to know is how does this change the workflow? How does it change turnaround time? How does it help with staff engagement?”

So far the instrument is delivering on the expectation of lowered turnaround times and making it possible for staff to be redeployed to other areas, Dr. Hansen says. “We think these advantages are generalizable across all different laboratory spectrums.”

TriCore Reference Laboratories has seen similar efficiencies with its use of WASPLab. “The workflow impact is huge, especially in such a strained environment that we’re experiencing now,” says Karissa Culbreath, PhD, D(ABMM), medical director of the Infectious Disease Department at TriCore. She shared her laboratory’s experience with WASPLab in an AI session last year at the AMP annual meeting and in a July 9 interview with CAP TODAY.

“What we’ve been able to see in our lab is that about 30 percent of urine cultures are generally negative and another 30 percent are normal urogenital cultures. Which leaves a small proportion that are positive and would require workup by the technologists,” Dr. Culbreath says.