Using Data Analytics and Machine Learning to Decode Cycle Times: How Infusion Centers Benefit

TOP - July 2020, Vol 13, No 4 - Value-Based Care

Every infusion center needs to understand what resources it has (or will have) available at any given time during any particular day—whether it is open chairs, treating registered nurses, or pharmacy staff. This knowledge is vital to decision-making that affects the patient experience, as well as staffing decisions, operations, and, ultimately, revenue.

Cycle times can be defined in different ways. For the purpose of optimizing chair resource utilization, they can be defined as the total time patients spend in an infusion chair, and they are a vital component of infusion center operations. Having a thorough understanding of in-chair cycle times can help level load the days and keep resources from being depleted or underutilized.

The Importance of Cycle Times

Infusion center appointments can range from a short 15-minute injection to a 9-hour lutetium Lu 177 dotatate (Lutathera) treatment. Every center has its unique mix of patients and associated treatments.

Budgeting patient hours accurately is key for the efficient utilization of resources; however, variance between the expected duration of the appointment and the actual time it lasts can be difficult to anticipate, because of patients’ reactions to drugs, difficulty with access, drug prep time, and sometimes even difficulty getting orders. When these durations are consistently different from what was scheduled, issues arise.

For example, if cycle times are longer than expected, unanticipated bottlenecks can arise from lack of chairs or nurses to accommodate all patients. This can lead to dissatisfied patients and frustrated nurses. Appointments may also run past regular operating hours or extend nursing shifts, which can increase pressure on the pharmacy to prepare drugs faster.

If cycle times are routinely shorter than expected, the center will be underutilized and treat fewer patients than its actual capacity. If only 4 appointments run 30 minutes less than expected, this would amount to 2 hours that could have been used. Also, patients who may want to come at a specific time are often turned away because schedulers recognize that time as booked, when it ultimately was not used. Similarly, nurses could be idle for longer than desired and have a crunch at other points in their week, making them wish for a more balanced schedule.

Diagnosing the Problem

Many reasons can cause cycle time to run long or short, which is par for the course in infusion. Some reasons stem from unpredictable individual level issues, such as bad reactions to drugs, difficult veins, the need for add-on blood or hydration, and more. Other reasons are linked to systemic issues, such as not including or misestimating prep time in the total chair time or treatment regimens that have been categorized incorrectly.

Although these problems are valid and common, accepting the inefficiencies they cause as the norm is not necessary. To fix issues properly, an infusion center has to conduct a root cause analysis, and here, data analytics prove invaluable.

There are many good starting points for examining why cycle times are consistently different from what was expected, but the recommended approach is to look at a typical time period and compare the expected duration to the actual cycle time for each patient. From there, results can be grouped into 3 categories—appointments that ran long, those that ran short, and appointments that are within range. Finally, these can be compared against several potential driving factors, such as time of day, day of week, the regimen, and the provider.

What the Data Can Tell Us

Let’s take a closer look at how tracking data can help to refine cycle time estimates. Looking at treatment length data over the course of a few weeks, a center may notice that more than half of its 2-hour appointments are running long (Figure 1). With this information, the center can dive in further to see if there is a subset of 2-hour appointments that may require a 3-hour block instead. Such an adjustment could have a significant impact on utilization outcomes.

Looking at cycle times by time of day can often help pinpoint opportunities to improve scheduling practices. Typically, infusion centers will aim to schedule their longer treatments earlier in the day to ensure they finish before close. Appointments that have more predictable cycle times should be scheduled closer to the end of the appropriate booking window, whereas those with greater variance should have more buffer time. Some simple guidelines for schedulers can be a factor in preventing nurses from staying late.

By looking at the day of the week, problem days may emerge (Figure 2). This could result from a specific regimen being more common on certain days, or clinical trials, which have a greater complexity and variability in cycle times, may happen on a specific weekday. Such an analysis helps identify the attributes of appointments that are consistently running long. Does a particular provider always work on those days? Do many patients need add-ons or fluids because of the treatment they receive on specific days? Answers to these questions should help determine the duration of the appointments needed.

Regimen analysis can explain whether a center is incorrectly scheduling appointments associated with a specific regimen; for example, if cisplatin and FOLFIRI appointments consistently run long, and paclitaxel appointments run short. Only by looking at the hard data do these issues tend to surface. This is vital information, because most centers have a cheat sheet for schedulers to know approximately how long they should schedule for each regimen. Cheat sheets need to be updated periodically, and data analysis can show why and how to do so.

Provider-based analysis (Figure 3) helps highlight which providers tend to over- or underestimate their treatment lengths. They may not be taking factors such as prep time into account in their orders. In addition, providers can have different methods for calculating treatment times. Understanding who is consistently out of range, then working on a standardized method for determining treatment duration is helpful.

These examples provide a snapshot of what can be learned through a focused analytical approach to infusion data. There are many more variables that should be considered if a center wants an in-depth picture of its cycle times.

Machine Learning Delivers Next-Level Insights

Many actionable insights can be taken from data analytics, but emerging technologies in advanced analytics and machine learning have the potential to take solutions to the next level.

By taking the input variables discussed here, any additional factors, and even unstructured notes, machine learning algorithms could be used to predict the most accurate expected duration for a given scheduled appointment.

Leveraging machine learning in this way would empower cancer centers to better keep their promise to patients about the length of their treatment and more efficiently use their resources. As schedulers book their patients’ appointment in real time, they could be made aware of the likely duration needed for that appointment based on the detailed information provided, not simply the treatment type. Machine learning is therefore an investment in the future that works for the good of patients, the staff, and the center itself.

Related Items
Universal CAR T-Cells a Promising Off-the-Shelf Therapy for T-Cell Acute Lymphoblastic Leukemia By Wayne Kuznar
TOP - July 2020, Vol 13, No 4 published on July 15, 2020 in Value-Based Care
ASCO Updates Its Oncology Medical Home Payment Model, Putting Patient Needs Up Front
TOP - July 2020, Vol 13, No 4 published on July 15, 2020 in Value-Based Care
Early Detection of Cancer Feasible with Blood-Based Tests
TOP - July 2020, Vol 13, No 4 published on July 15, 2020 in Value-Based Care
Real-World Healthcare Utilization and Costs Support Broader Use of CAR T-Cell Therapy
Phoebe Starr
TOP - March 2020, Vol 13, No 2 published on March 11, 2020 in Value-Based Care
First National Study: Geographic Socioeconomic Disparities Mirror Substantial Gaps in Survival in Children with AML
Chase Doyle
TOP - March 2020, Vol 13, No 2 published on March 11, 2020 in Value-Based Care
Real-World Applications of Value-Based Care in Oncology
TOP - March 2020, Vol 13, No 2 published on March 11, 2020 in AVBCC 2019 Highlights, Value-Based Care
Last modified: July 30, 2020