Newsbytes

Editors: Raymond D. Aller, MD, & Dennis Winsten

Leveraging lab analytics: from requesting data to implementing tools

January 2024—When the medical microbiology laboratory at Yale-New Haven Hospital makes operational changes, it uses data analytics to monitor their impact. Yet the process of implementing laboratory analytics can be challenging.

“I’ve come across bad data sets, incomplete data, and situations where a simple data query is not going to be adequate for addressing the question at hand,” says David Peaper, MD, PhD, director of Yale-New Haven’s microbiology lab and associate professor of laboratory medicine at Yale School of Medicine.

As a microbiologist without formal training in data science, Dr. Peaper developed and honed his technical skills and techniques for laboratory data analysis on the job while creating and implementing data-driven tools for assessing lab operations. The lessons he learned in the process, sometimes by trial and error, have provided insight into Yale-New Haven’s laboratory operations.

“We are subject matter experts in our disciplines, so we need to develop data-extraction and -analysis skills because we are best suited to have at least this basic understanding of what is going on in our laboratory,” Dr. Peaper said during a presentation about creating a laboratory analytics program, conducted at the 2023 annual meeting of the Association for Diagnostics and Laboratory Medicine.

During the planning phase of such undertakings, the laboratory should develop a structured way to ask data analysts for information that will serve as the basis of the project, Dr. Peaper explains.

Using turnaround time as an example, he says the laboratory must determine what aspect or portion of TAT it wants to analyze using this common metric for evaluating lab efficiency. Turnaround time could refer to the preanalytical part of the process, stretching from when a test is ordered until it is collected. It could also refer to the time from when a specimen is collected until it is received in the lab or the time from specimen collection until results release. Therefore, the laboratory should include in its data request the time stamps that define the portion of the turnaround time process that it will analyze and, more importantly, why it wants to analyze that portion of the TAT process, Dr. Peaper says.

“Maybe you are really interested in the time from specimen receipt until Gram stain for the pediatric emergency department or the pediatric ICU,” he adds. “That is another level of detail that you need to convey.”

Sharing details is important in part because most data analysts do not have deep technical knowledge of lab practices, Dr. Peaper says. Therefore, they may not know, for instance, that microbiology laboratories often receive a specimen, stain it, and produce an initial result within an hour but that it may take anywhere from a day to a few weeks to receive additional test results. Consequently, “the more specific and detailed your request for information can be, the more likely you are to get useful data earlier.”

Sometimes laboratories need data on patients who are located in a particular building or floor in the hospital complex, he notes. If the subunit of pediatric patients that the laboratory is studying is housed in North Pavilion 12, for example, providing that specific location could help data analysts verify that they are retrieving the desired information.

“There is a tendency to say, ‘Just give me everything and I’ll sort it out,’” Dr. Peaper says. But that approach is not only less efficient but less safe, he adds. Most laboratory data contain protected health information, and receiving a larger-than-needed data file could potentially put a greater amount of patient information at risk of being compromised in the event of a cyberattack.

Once the lab receives the data it requested, Dr. Peaper recommends double- and triple-checking it for accuracy. “No one knows your lab data like you,” he says.

In reviewing data files obtained for analytics purposes, Dr. Peaper has found data that are missing critical elements, such as birth dates, and uncovered proficiency test data integrated with regular test results. He has also seen data with impossible values, such as results for a 30-minute test that have time stamps suggesting the test took far shorter or far longer than the expected time.

Depending on the intended use for the data, Dr. Peaper may exclude contaminating data or data with impossible values from further analysis. He recommends carefully documenting which data are being excluded, in the same way data parameters are defined in academic research papers.

Whether the data are to be presented “to my laboratory team, or my department chair, or external stakeholders, the more formal the process you use, the better,” he says.

Dr. Peaper initially used tools embedded in Microsoft Excel for analyzing data extracts, which, he says, works well for ad hoc analysis.

Comma-separated values files or Excel files can be dropped into Excel quickly, and the program’s pivot tables make it easy for users to sort, group, and rearrange data for analysis. However, Excel has limitations, he notes. For example, the program is capped at 1,048,576 rows of data, and the workflow processes for working with data are manual.

One project that Dr. Peaper says “bumped up against the limits of what is reasonably accomplished in Microsoft Excel” involved assessing the clinical and laboratory impact of a new polymerase chain-reaction testing protocol for the flu. The laborious project linked laboratory report data with clinical emergency department report data via encounter IDs to evaluate the impact of the testing protocol on time to PCR result, length of emergency department stay, and time to Tamiflu administration. The project demonstrated that the new protocol resulted in faster PCR result TAT, shortened emergency department stays, and a shorter time to Tamiflu prescription.

“We’re talking about linking multiple events, including clinical parameters, in a particular encounter across time, and there is a tremendous opportunity for error,” Dr. Peaper says, in explaining the technical challenges of the project.

To address the difficulties of working with multiple sets of data and increasingly sophisticated rules, Dr. Peaper learned Python. This was a worthwhile undertaking, he says, because “when you start trying to combine data or filter data, there is a lot of opportunity to corrupt your data. The more you do things manually, the more likely this is to happen.”

With Python, rules, such as for exclusion criteria, are written in code, which not only automates processes but makes workflows easier to adjust. Therefore, if a laboratory decides to change its exclusion criteria for a project, anyone with knowledge of Python can alter that part of the code related to exclusion criteria. Making that type of change in Excel would be tantamount to starting over, Dr. Peaper says.

While Dr. Peaper acknowledges that the learning curve can be long for determining how to logically ask questions of data for coding purposes, he considers Python to be a valuable tool for pathology laboratories.

“By combining our subject matter expertise as laboratorians with being able to extract and analyze data, we breed collaboration with our colleagues and our institutions,” Dr. Peaper concludes. “These collaborations, and the insights that lab data provides, can help pathologists demonstrate their value to their institutions.”

—Renee Caruthers

Secondary uses of data focus of Joint Commission certification program

The Joint Commission has introduced the voluntary Responsible Use of Health Data certification program for U.S. hospitals and health care organizations.

The program will provide guidance on how to safely use data for purposes other than clinical care, such as for improving quality and operations, developing algorithms, and advancing artificial intelligence. It will also recognize health care organizations that establish policies and procedures to protect health record data.

“The certification will provide an objective evaluation as to whether an organization is committed to utilizing best practices in its secondary use of data and promoting responsible use of data,” according to a press release from the nonprofit standards-setting and accreditation organization.