---
name: infusion-kpi-review
description: Reviews an infusion center's monthly KPIs (chair utilization, no-show rate, referral to first treatment time, referral conversion rate and claim denial rate) from de-identified or aggregate report exports the user pastes or uploads. Use when an infusion operator, CFO or analyst asks for a monthly numbers review or asks why one of these KPIs moved.
---

# Infusion KPI Review

Review an infusion center's monthly numbers from exported reports. Check that each KPI is defined the same way every month, compute it, and explain what moved.

## Privacy first: no PHI

Say this to the user before they share anything:

> Please do not paste or upload patient names, dates of birth, medical record numbers, addresses, phone numbers, insurance member IDs or any other detail that identifies a patient. Use de-identified or aggregate exports only: counts by site, day, payer or drug, or rows with a random ID in place of the patient.

If the user shares a file that contains identifying details anyway:

1. Stop and tell them what you saw (for example, "this file has a patient name column").
2. Do not repeat, quote or summarize the identifying values.
3. Ask them to remove those columns and share it again, or to share aggregate counts instead.

Never ask for identifying details to "match" records. Ask for an order ID, appointment ID or a de-identified key instead.

## Step 1: Agree the definitions

Write each KPI down with what you count, what you divide by, the grain of a row and the clock. Confirm these with the user before computing. If they have used a different rule in past months, keep their rule and note it, so months stay comparable.

| KPI | Formula | Grain | Clock |
|---|---|---|---|
| Chair utilization | Chair time used ÷ (staffed chairs × open hours × operating days) | Chair × day, rolled up to site × month | Inside each site's open hours, local time |
| No-show rate | No-shows ÷ appointments scheduled for the day | Appointment | Appointment date |
| Referral to first treatment | Median and 90th percentile days, order received to first completed administration | Order | Order received to first administration |
| Referral conversion rate | Orders reaching a first completed treatment within a fixed window ÷ orders received that month | Order cohort by month received | Order received date plus the window |
| Claim denial rate | Claims, lines or dollars denied on first adjudication ÷ claims adjudicated for the first time | Claim or claim line | Remittance date |

Questions to ask:

- **Chair utilization.** Use completed infusions only: completed visits times average door-to-door chair time (vitals, pre-medication, infusion, observation), or the scheduled length of each completed visit if that is all the export has. Do not count booked appointments: no-shows and same-day cancellations fill the schedule, not the chair. If the user only has booked time, report it separately as "booked chair time" and do not call it utilization. Count staffed chairs, not every chair in the room.
- **No-shows.** Which statuses count as no-show and which as same-day cancellation? Keep them as two rates. Leave advance cancellations out of the denominator. Do not count a rescheduled visit twice.
- **Referral to first treatment.** Does the clock start when the fax arrived or when the order was entered? A booked appointment does not stop the clock; only a completed administration does.
- **Referral conversion.** What is the follow-up window? Orders still open at report time are open, not lost.
- **Denial rate.** Claims, lines or dollars? Initial (first adjudication) or final? Normal adjustments are not denials: leave out the contractual write-down (reason code 45), Medicare sequestration (reason code 253) and patient responsibility.

## Step 2: Check the exports

For each file, confirm what one row is and which date it uses. Common sources:

- **Appointments**: status, location, chair, scheduled start and end, appointment date.
- **Orders**: received date, referring office, drug, payer, including orders with no appointment yet.
- **Administrations**: completed visits where drug was given.
- **Remittances (835) or a denial report**: from the billing system or clearinghouse, with reason codes and amounts. Infusion software often holds expected amounts, which are estimates, not payments.
- **A chairs and hours table** for each site.

Checks before computing:

- A visit counts as completed only when the patient came and drug was given. A check-in with no drug is not a completed visit.
- Referral KPIs must start from orders, not appointments. A report built from appointments cannot see orders that never got one, which are the stuck ones.
- One order has many appointments. Make sure an order-grain KPI is not counted once per appointment.
- If several sites are included, keep site as a column and use each site's local time.

## Step 3: Compute

- **Rates across sites:** add numerators and denominators across sites, then divide. Do not average site percentages; a small site would get the same weight as a large one.
- **Referral to first treatment:** report the median and 90th percentile, not the average. Also show how many orders are still waiting, by age, so the slowest cases do not drop out. If stage dates exist (benefits verified, authorization submitted and decided, scheduled), report each stage on its own; stage medians do not add up to the total median.
- **Referral conversion:** follow each month of referrals as its own cohort. Do not divide this month's first treatments by this month's referrals. Show converted, lost (with loss reason) and open. Stage losses multiply, they do not add.
- **Denial rate:** show both count and dollars. Group denials by reason, payer, drug and site. Keep prior authorization denials separate from claim denials.
- **Recent months move.** Referral conversion and denial rates for the latest month are not final. Mark them as provisional.

## Step 4: Review

For each KPI, compare with the prior month and, if available, the same month last year or the user's baseline. Then:

1. Say whether it moved, by how much, in the units the user uses.
2. Check whether the move comes from the data before calling it real: a change in status coding, a new site, a billing entity change that makes a site appear twice, a late export, or a different definition.
3. Name the most likely operational driver the data supports, and the one cut that would confirm it (by site, weekday, payer, drug or stage).
4. Read KPIs together. A fast referral median with falling conversion means referrals are being lost. Low utilization with a full floor can mean nurse staffing or pharmacy prep is the limit, not chairs.

Do not compare with industry benchmarks. Published infusion benchmarks are rarely defined the same way. Compare the center with itself and help it set a baseline.

## Output format

1. **Data note**: files used, grain, period, sites, and a line confirming no identifying details were used.
2. **Definitions used**: the agreed rules from Step 1, with any assumptions marked.
3. **KPI table**: KPI, this month, prior month, change, numerator and denominator, provisional flag.
4. **What moved and why**: one short paragraph per KPI that changed, with evidence.
5. **Data issues**: anything that could make a number wrong.
6. **Questions for next month**: the cuts or extra fields that would answer what is still open.

Built by Greyfield Data (greyfielddata.com). Free to use and adapt.