Technology
Capillary Health maps where care is hard to reach, tests plans on simulated regions, routes home visits on real roads and checks staff schedules for rest. Scheduling is live; the maps and routing run as demonstrations on public and synthetic data.
Map
The access map scores all 3,144 U.S. counties and county equivalents for how hard long-acting HIV care is to reach. A county’s score is the geometric mean of its HIV burden percentile and its access gap, so a county scores high only when both are high. 629 counties are modeled as access deserts.
Data
County scores from MapMedicine’s public HIV access map, built from AIDSVu county HIV prevalence (2023), HRSA Ryan White HIV/AIDS Program sites (2023) and U.S. Census county boundaries.
The score is a research signal for planning, not a clinical recommendation.
Access to long-acting HIV care, by county
Public county data, 2023.
A map of 3,144 counties
The map draws every county and county equivalent in the 50 states and the District of Columbia. It loads about 1 MB of county shapes.
County figures are public aggregates. A mapped site is a Ryan White program site that could plausibly offer long-acting care. The map does not verify a product at any site.
Model
The benchmark asks whether planning long-acting HIV visits across providers covers more visits than each provider planning alone. Four planning rules ran against identical resources in 7 states, 500 simulated scenarios each, 3,500 in all.
Against a planner that schedules one provider after another and then repairs the plan, Capillary tied in 95.6% of scenarios, came out ahead in 4.4% and fell behind in none.
The prespecified rule for an advantage was not met: the median paired difference was 0.0%.
In the 1,869 scenarios where providers could share capacity, planning across providers covered more of the doses whose dosing window closes within three days.
+12.8%
median coverage of those doses (mean +19.3%)
A subgroup of the 3,500 scenarios, against a weaker comparator than the result beside it.
Against the well-run planner, the difference is exactly zero when nothing is shared, and grows as more pairs of providers allow sharing.
These are modeled planning results on synthetic delivery scenarios. They are not clinical outcomes, provider acceptance or adoption evidence. Read the methods and data.
Four working demonstrations on public county data and synthetic operations. No account needed.
Route
The routing engine plans visits for clinicians, couriers and equipment. Google OR-Tools builds a first plan that respects each person’s skills, shift, lunch break and visit limit. Then adaptive large neighborhood search takes parts of the plan apart and rebuilds them, up to 10,000 times, keeping the best plan it finds. Drive times come from OpenStreetMap roads, through the OSRM routing engine.
The solver gets coded IDs and map coordinates, never names or medical record numbers.
What a plan is scored on
Schedule
Scheduling is live. Capillary checks every assignment as it is saved. Each clinician’s own hours page marks their hardest shifts, and department analytics reads a group’s calendar for patterns.
Each save checks role, overlapping shifts, minimum rest and approved time off for every assignment in the month, including the person’s published shifts at other sites in Capillary. Rest is counted in elapsed hours, so it stays right when daylight saving time starts or ends.
Messages a save shows, with sample shifts and dates.
From shift times
Flags from the shift times alone: rest under 11 hours, a backward rotation, a 4th day or 3rd night in a row, or a shift over 12 hours. Two or more flags mark a shift high burden.
Risk model
A risk index after Folkard and Lombardi: night shifts and long shifts carry more risk, and the risk rises with each night in a row.
Alertness model
The three-process model of alertness, run over the 48 hours before a shift, with sleep assumed from the shift times.
The models give estimates about people in general, not a measure of one person. Capillary does not measure sleep.
Successive night shifts
Successive day shifts
From a group’s calendar, department analytics counts hours, nights, weekends, holidays and rest for everyone on it, and shows coverage by hour of the week. It also learns the shifts each colleague usually works, so the swap finder lists likely colleagues first.
A pattern such as “nights only” is flagged only when chance alone would produce it less than 1% of the time, and never for someone with fewer than 8 shifts.
Evidence
Templates, rule checks, publishing, requests, trades and sick calls.
From a person's own calendar link or their group's calendar.
Public county data; no patient data.
The routing engine on synthetic visits, clinicians and couriers.
Synthetic delivery scenarios. Not clinical outcomes or adoption evidence.
The routing engine follows the home health care routing and scheduling problem as the operations research literature defines it.
Book a call with a logistics expert. Bring your schedule, your region or your fleet, and we’ll show you what Capillary does with it.