Clinics carry a quiet arithmetic. Each provider has minutes to spend, rooms have to turn over, and patients arrive in patterns that feel random but are not. When the numbers look off, leaders often reach for averages and percentiles. Then they wonder why that “typical” 22-minute visit hides hour-long waits, why some days feel calm while others melt down, and why overtime creeps in even when schedules look reasonable on paper. The answer often sits in the shape of the data, not the average. In ambulatory operations, throughput metrics are frequently bimodal. If you do not recognize the second peak, you keep chasing the wrong cause.
I learned this the hard way directing an internal medicine clinic with 22 providers across two sites. Our patient satisfaction scores had a stubborn dip tied to wait time and perceived chaos. We celebrated when the average cycle time dropped from 78 to 63 minutes in Q2, then got hammered by complaints about two-hour delays in Q3. The overall average looked fine. The distribution, once plotted, told the real story: a clear bimodal chart for both arrival-to-room and cycle time, with one cluster around 45 to 60 minutes and another sprawling between 100 and 140 minutes. Two distinct experiences were happening under the same roof. We were measuring one clinic, but running two.
What “bimodal” looks like in clinic throughput
A bimodal distribution has two peaks. In clinic throughput, you tend to see it in measures like:
- Arrival to triage or room start. Room start to provider in. Provider in to visit end. Lab or imaging turnaround. Full cycle time, door to door.
When you graph these times as a histogram across enough days, the curve does not form a single hump. Instead, it shows one cluster of visits progressing smoothly and another cluster stalled by a shared set of conditions. The first cluster often represents routine, well-matched visits during normal staffing and room availability. The second exposes bottlenecks: demand surges, provider or room mismatches, lab congestion, late-day backlogs, cross-coverage interruptions, or coordination steps like immunizations that appear trivial until you watch them contribute five minutes at a time.
If you do not look for the second peak, you may rationalize the variation as “complex patients” or “no-shows on a rainy day.” Those explanations carry some truth, yet they miss structural drivers. When two peaks persist for weeks, something systematic is splitting your flow into two streams.
Why bimodality emerges in clinics
Healthcare is loaded with arrival variability and service variability. Toss in limited capacity that cannot flex easily hour to hour, and you get nonlinear effects. Several patterns create a second peak:
- Scheduling compression: Stacking the morning with short visits and the afternoon with long or procedure-heavy visits increases the chances of afternoon pileups. The promise to “catch up at lunch” rarely survives walk-ins and precepting. Resource synchronization failures: A provider is ready, but a room is not. A room is ready, but a nurse is drawing blood in the next bay. A patient is ready, but a vaccine needs a thaw or a prior suppression workflow in the EHR. Small misalignments compound. Late-day demand cliffs: Walk-ins, urgent add-ons, and late arrivals cluster in late afternoon in many clinics. If providers are fixed in number and rooms are saturated, the queue expands. Even one prolonged procedure can spill into the next hour, creating a cascade. Specialist sharing: In multispecialty sites, a single respiratory therapist, point-of-care lab machine, or interpreter can become a gate. Everything runs fine until multiple services need the same resource at once. EHR documentation debt: Beginning-of-session progress is swift, then chart lag builds. When documentation spills to the end of visits, checkout stalls, MA coverage thins, and the hallway backs up. Asymmetric staffing: Coverage looks balanced on a schedule grid, yet skill mix is not. Three newer MAs paired with complex providers will drift into the second peak even if headcount is full.
Bimodality is not only about “bad days.” It often results from intentional design that optimizes for a wrong target, for example maximizing provider template utilization while ignoring the micro-queues the template creates downstream.
Finding the second peak in your data
Most EHRs can export timestamps for arrival, room start, provider in, provider out, orders complete, and checkout. Pull a month or two of data, exclude extreme outliers if they are data errors, then make simple histograms for each segment and for total cycle time.
Look for a valley between two clusters. If you have the tools, fit a two-component Gaussian mixture model to quantify the peaks. But you do not need statistics class flashbacks to see the shape. A basic bimodal chart often convinces teams more than a table of averages ever will.
Then segment. Plot separate histograms by:
- Time of day block: opening, mid-morning, lunch hour, mid-afternoon, last hour. Visit type: acute, routine follow-up, annual, chronic disease check, procedures. Provider: especially those supervising learners, or with different panel complexity. Room or pod: clusters often map to physical space. MA or RN assignment and coverage patterns.
The trick is to avoid overfitting. You are not hunting for the perfect subgroup. You are looking for consistent contexts in which the second peak grows. In our clinic, the late-afternoon block, chronic disease visits with labs, and a specific two-room pod showed an exaggerated second peak. The cause was not one villain, but a tangle: lab courier timing, room turnover habits, and a heavy template for a senior provider who never ran behind until three unavoidable teaching moments stacked.
What averages hide and why that matters
Operational decisions based on averages are like tailoring a suit using your height alone. The average cycle time, average patients per hour, and average room turns have their place, but they smooth away queue dynamics. Three specific risks follow:
- False reassurance: An average of 65 minutes masks that a third of patients wait 110 minutes. Leadership accepts this as “within target,” and front-line staff absorb patient frustration without air cover. Misplaced fixes: You run generic throughput kaizens that beat up on triage or intake, because that is where everyone sees the line, while the real cause sits with lab choke points or vaccine workflows later in the visit. Inequity: Some patients, often those scheduled late in the day or requiring language support, disproportionately fall into the second peak. If you do not see the bimodality by segment, you miss an equity issue disguised as operations.
The fix starts with measuring and visualizing distributions, not just central tendencies. Then you make targeted changes to collapse the second peak toward the first.
Practical levers that collapse the second peak
There is no universal recipe. You combine several changes aligned to your particular drivers. The goal is to decouple the conditions that create runaway queues from the rest of your flow and to absorb variability earlier with less damage.
Rebalance appointment types to level the day. If your grid loads long visits after noon, you are choosing a predictable surge. Mix complex visits across blocks, or create two protected long-visit slots in the morning. In a family medicine site I supported, rebalancing reduced the late-afternoon second peak by about 25 minutes without altering total volume.
Align rooms per provider to visit length variance. Providers with high variance in visit duration benefit from an extra room, not to see more patients, but to absorb stochastic spikes. Conversely, low-variance templates can run cleanly on fewer rooms. We moved one high-variance provider from two to three rooms and trimmed their tail of 120-minute cycle times by half without changing their template.
Batch-sensitive steps earlier. Anything that triggers a bottleneck later should be pulled forward: labs ordered pre-visit with standing orders, pre-visit questionnaires completed online, vaccine insurance checks done at check-in, imaging scheduled ahead when indicated. This turns lumpy, late-day work into early, predictable prep. For prediabetes group visits, final A1c draws at the end always created a lab queue. Pulling those to pre-visit cut post-provider idle time by 12 to 18 minutes per patient.
Create a rapid-turn track inside the same clinic. Not a new urgent care, just a deliberately designed path for specific visit types that seldom need the full room-provider cycle. Suture removal or blood pressure rechecks can be nurse-driven visits in flex rooms with standing protocols and brief provider sign-off. The value is not the saved minutes per quick visit. It is the congestion avoided in the general queue.
Use time-of-day staffing flexibility. Fixed FTEs do not have to mean fixed half-hour starts. Stagger MA or RN start times so that the ratio of staff to expected arrivals matches demand shapes. In pediatric practices, early mornings and late afternoons often carry peaks. Sliding two MAs 30 minutes earlier and two 45 minutes later cut visible hallway queues much more than adding a single float who arrived at 10 a.m.
Right-size and standardize room turnover. The second peak often correlates with rooms that cannot be reclaimed quickly because supplies are inconsistent or the order of reset varies by person. We color-coded drawers for standard setups and placed a simple one-page reset checklist inside each cabinet. Average room turnover time dropped only four minutes, but the variance tightened dramatically, and the afternoon peak softened.
Engineer documentation to stay within the visit. When notes spill to end-of-session clean-up, support staff drift to help elsewhere, and checkout jams as final orders or notes delay disposition. Tighten templates, outsource certain note elements to scribes or voice capture judiciously, and adopt “close the note before leaving the room” for eligible visits. Providers will push back if quality suffers; trial for specific visit types and measure both time and accuracy.
Protect block time from add-on bleed. Add-ons will always happen, but they can be channeled into short, pre-labeled slots or a specific provider’s access block to avoid fragmenting the entire template. If your data show that a single unplanned 40-minute add-on at 3:30 p.m. triggers a 90-minute second peak, create rules that shift those to a 2:00 p.m. flex or next-morning access.
Rethink lab and imaging couriers. In multi-tenant buildings, courier pickups and imaging access windows often shape your second peak by accident. If the last lab courier leaves at 4:15 p.m., a lot of blood draws get rushed between 3:30 and 4:10, queuing phlebotomy and slowing room turnover. Negotiate a 4:45 pickup or split pickups to 3:30 and 5:00 twice a week and watch the afternoon distribution tighten.
Sharpen handoff protocols. A striking number of “mystery” delays come from ambiguities at transitions: who rooms the patient after a stat EKG, who tracks down an interpreter, who escorts to imaging. Write short, visible rules and place them where decisions occur. Each handoff saved might be two minutes, but you are shaving the variance, not just the mean.
Using a bimodal chart to steer conversations
The politics of clinic flow can be delicate. Providers do not want to feel blamed. Nurses and MAs are exhausted with meetings about “efficiency.” Administrators get stuck between patient complaints and budget limits. A simple, clear bimodal chart often becomes the neutral artifact that re-centers the team around the shape of the problem.
Bring three visuals to your next huddle:
- The overall cycle time histogram for the last six weeks, clearly showing the two peaks. The same histogram split by time of day or by visit type, whichever exhibits the starker divergence. A cumulative distribution plot that shows, for instance, that 30 percent of patients cross the 90-minute mark on Tuesdays after 3 p.m., versus 8 percent in mornings.
Then speak in concrete trade-offs. We can move two complex visits from afternoon to morning per provider and reduce the late-day second peak by 20 minutes, or we can add one MA from 2 to 6 p.m., which costs X and buys Y minutes. Set a target on the second peak height or location, not just the average. People understand peaks and tails when they see the curve.
The value of small experiments
Flow changes land better when you agree to test, not to overhaul. Use a tight driver diagram, pick one or two changes, and run a two-week Plan-Do-Study-Act cycle. Keep measurement simple: daily median cycle time, the proportion of visits over 100 minutes, and a qualitative heat map by hour. If you have the analytic support, track the two peaks’ centers and proportions.
In a cardiology clinic, we piloted three moves: pulled EKGs earlier, staggered MA starts by 30 minutes, and reserved two 20-minute mid-afternoon “shock absorber” slots for urgent learn six sigma add-ons. After 10 clinic days, the late peak’s median dropped from 118 to 92 minutes, and its share of total visits fell from 37 to 24 percent. Volumes held steady. Staff reported the hallway “felt” calmer, which matters more than leaders admit because calm reduces error pressure and burnout.
Another site tried to fight the second peak with a lunch huddle and generic reminders to “stay on time.” Nothing changed. Their second peak was rooted in respiratory therapy availability and a single PFT room shared across three providers. Redistributing PFT slots and moving one provider’s complex COPD follow-ups to mornings did more in a week than six months of exhortations.
The hidden role of variability and buffers
Queueing theory explains much of what we observe. When utilization gets high and variability is not managed, waits explode. In clinics, you cannot hold a huge inventory of empty rooms and idle staff to buffer every surge. You can, however, reduce arrival variability and service variability, then add small, well-placed buffers.
Reduce arrival variability by smoothing templates, redirecting walk-ins to designated windows, confirming and pre-scheduling labs or imaging, and offering narrow same-day access blocks that you fill predictably. Reduce service variability by standardizing intake, right-sizing room setups, and creating clear exception paths for complex visits.
Buffers do not have to be people. They can be time blocks, flex rooms, dedicated quick-turn bays, or even technology that decouples steps, such as allowing vaccine prep in a separate area before the room is free. The key is deliberate placement. A buffer added before the main bottleneck can absorb random spikes; a buffer after the bottleneck only hides pain without reducing it.
When the second peak is unavoidable
Not every second peak is a pathology. Some clinics serve populations with genuine late-day access needs due to work schedules or transportation. Pediatrics sees after-school spikes; OB often handles late ultrasounds and NSTs to match maternal needs. The goal shifts from collapsing the second peak to containing it humanely.
Strategies include dedicated evening teams calibrated for the surge, explicit communication about expected waits with visible progress boards, amenities that matter during longer stays, and active monitoring to escalate when the queue slips beyond thresholds. In a pediatric practice, we staffed a 3 to 7 p.m. crew with a different skill mix, added a playroom attendant, and displayed a simple “your visit is in stage 2 of 4” indicator. Families felt informed, and even when waits ran 20 minutes longer, complaints dropped.
Data quality traps that mimic bimodality
Before you act, confirm that the second peak is not an artifact. Timestamps sometimes reflect documentation habits rather than actual events. Two common pitfalls:
- Provider “in” time entered in batches at the end of hour blocks: This creates a false second peak near 60 minutes. Audit a sample through direct observation or badge swipe logs if available. Checkout timestamp missing until labs finalize: Some EHRs record visit end only after ancillary orders are signed off, which can inflate cycle time for certain visit types. Separate clinical end from administrative close to understand true patient experience.
Clean your definitions, retrain on timestamp accuracy where needed, and re-plot. The real second peak will persist.
Equity and the second peak
Bimodality can reflect inequity disguised as logistics. Patients needing interpreters, those relying on public transportation, and those with lower digital access for pre-visit tasks are more likely to land in the slower stream. If late-day slots are all that remain when call center scripts prioritize early blocks for online self-schedulers, you are sorting people by resource access, not by need.
Audit distributions by language preference, insurance type, and scheduling channel. If the second peak concentrates in these groups, your fixes should include fairer slot allocation, interpreter availability aligned with demand, and offline alternatives for pre-visit completion. Equity-aware throughput is not charity. It prevents rework, missed labs, and angry revisits that burden the system later.
Measuring what success really looks like
Sustained improvement does not mean pushing the whole curve left at any cost. A healthier distribution has:
- A smaller or eliminated second peak. A tighter spread around the primary peak. Fewer extreme outliers on the right tail. Stable or improved quality metrics and staff well-being.
Track patient “time in stage” alongside overall cycle time. If arrival-to-room improved but provider-in-to-visit-end ballooned, you moved the queue, you did not resolve it. Monitor same-day overtime minutes and end-of-day “last patient out” time. Those metrics often mirror the second peak’s behavior.
Listen to staff. The floor knows where the day buckles. In one clinic, an MA described their 4 p.m. pattern as “we sprint, then we stand,” which mapped perfectly to a shared spirometry device opening then closing around the respiratory therapist’s break. A small schedule tweak flattened the sprint-stand cycle.
An anecdote from the field
Several years ago, a multispecialty clinic asked for help because their net promoter score tanked on Thursdays. Their averages sat within target. The histograms, though, showed a giant late-day second peak on Thursdays only. Thursdays hosted a rotating sports medicine clinic, and the physician liked to keep post-procedure observation patients in two rooms near the checkout desk so they would be easy to see. The observation policy required vitals every 15 minutes for an hour, and the MA assigned to that pod was also covering injections for primary care next door.
No one had stitched the pieces together. The sports clinic ran fine. Primary care looked fine on averages. Together, they created a slow-moving jam every Thursday at 4 p.m. The fix was embarrassingly straightforward once visible: move post-procedure observation to a nook near the nurse station with a clock and supplies, assign a dedicated MA during that two-hour window, and pre-block a single injection slot per 15 minutes in an adjacent pod. Thursday’s second peak shrank by 35 minutes within two weeks, and the staff reported the tension in the last hour dropped significantly.
Bringing it all together
Bimodality in clinic throughput is a signal, not a curiosity. It points to two distinct operating states coexisting in your day: a flow that works and a flow that does not. The job is not to lecture people into moving faster. It is to redesign the system so the second state cannot form easily, or when it does, it stays bounded and humane.
That work begins with drawing the curve. Put a simple bimodal chart on the wall and let people tell you what they see. Segment by time, visit type, and space. Test small changes aimed at the drivers that inflate the second peak. Respect the edge cases where access realities create a necessary late-day surge, then support that surge deliberately. Keep your measures honest and your definitions clean.
Most of all, avoid the lure of averages as your north star. Patients do not experience an average. They experience the state of the system they walked into. If you make the worse state rare and less punishing, the whole operation feels different, even if the mean hardly moves. That is the paradox of throughput improvement in healthcare: shaping the distribution often beats chasing the number.