Automated dispensing cabinets (ADCs) have become commonplace in healthcare environments, ensuring timely access to a range of medications near the point of care. For busy clinical pharmacy teams, ADCs can help minimize medication waste, ensure medication security and accountability, and alleviate some of the workload of managing a decentralized inventory.
But without a sound fulfillment strategy, ADCs can pose logistical challenges. Running out of medication, expired medications, and frequent refills are all a drain on pharmacy resources.
A Dec. 8 session at the Midyear Clinical Meeting & Exhibition shared a case study of how the pharmacy informatics team at University of California San Francisco (UCSF) Health optimized its ADC management strategy, ensuring the right medications are on hand when needed with the fewest refill trips.
Optimizing Automated Dispensing Cabinet (ADC) Inventory: A Data-Driven Approach Using Linear Programming described UCSF Health’s transition from a cart-fill dominant model to using ADCs in 2021. Presenters noted that cart fills are typically used for stable, scheduled, patient-specific doses and high-cost therapies, while ADCs are useful for storing fast-moving and urgently needed medications, as well as controlled substances. Hospitals and health systems may employ a blend of ADCs and cart fills.
Today, about 85% of UCSF Health’s inpatient-administered medication comes from ADCs, said Chih Hsu, director of pharmacy. As the health system prepared to acquire three new hospitals in 2024 and 2025, roughly doubling the ADCs the pharmacy would be responsible for, there was a need to make provisioning all those machines more efficient and effective.
“Leaving the pharmacy once to refill 20 medications, that makes sense. But if the pharmacy technician is making three or four refill trips throughout the day, that’s time taken away from doing other work,” Hsu explained.
First, UCSF Health removed rarely ordered medications to free up valuable space in the ADCs. The pharmacy then adjusted the minimum and maximum stocking requirements for certain drugs to reduce the number of refills.
Further optimizing the medication inventory would require more complex calculations. A health system can have hundreds of ADCs that feature different storage capacities and unique arrangements of pocket types.
UCSF Health turned to a mathematical technique called linear programming to design a tool that automates the selection of ideal ADC configurations. Given each medication, its dispensing rate, and all possible storage options, the model can determine optimal ADC pocket-size assignments. The tool can even recommend which medications to place in empty pockets.
The pharmacy team decided that refill trips were the most appropriate metric to optimize. “It's a clear, logical, and attractive goal to reduce that refill frequency. We won’t get to zero, but we can get as low as possible,” said data science pharmacist Craig Michael.
The model was named Timbuk2 after the popular San Francisco-based gear company, a nod to the so-called knapsack problem often used to teach linear programming. Optimizing an ADC is much like choosing what supplies to carry in a camping backpack, given certain space and weight constraints, Michael explained.
He walked Midyear attendees through an exercise to optimize three medications with different packaging sizes and dispensing rates: acetaminophen 500 mg tablets, divalproex 250 mg tablets, and hydrocortisone 100 mg vials. Of the six possible configurations across three ADC pocket types, the best choice — that is, the one requiring the fewest refill trips — placed divalproex in the smallest pocket, acetaminophen in a medium-sized pocket, and hydrocortisone in the largest.
Such small-sample calculations can be done manually in minutes. But for an ADC carrying 100 to 200 medications, “working through that in Excel is a non-starter,” Michael pointed out. “That’s exactly why linear programming exists.”
The Timbuk2 pilot achieved exactly what UCSF Health hoped: After 60 days of implementing its recommended changes, they reduced ADC pocket refills by about 57% and refill trips by 36%.
Linear programming models have their limitations, Hsu and Michael noted. For example, Timbuk2’s calculations represent medication usage at a single point in time and cannot easily adapt to seasonal fluctuations or unexpected surges in demand.
UCSF Health plans to refine Timbuk2 to account for matrix drawers, or ADC pockets with adjustable subdividers. They will also investigate the tool’s cost savings, impact on staff, and long-term time savings, then look to implement it across the entire health system. Timbuk2 could even determine the ideal pocket combinations for new ADCs.
The speakers encouraged health-system pharmacists to adapt the linear programming approach to the needs of their patient populations, budget, and supply chain limitations.
“You want to make sure your ADC dispensing model makes sense,” Hsu said.
