# Census Analysis > Turns raw SNF census data (daily/midnight census exports or hand-typed numbers) into an operational census brief: occupancy, ADC and payer mix, skilled mix, admission/discharge flow, length of stay, trends, red flags, and recommended actions. Use whenever the user mentions census, occupancy, payer mix, average daily census (ADC), skilled mix, bed management, midnight census, admissions/discharges pipeline, backfill, bed holds, or pastes or uploads a census spreadsheet. Also use for "how full are we," "are we losing Medicare days," or any request to analyze building occupancy trends. - Skill: `magicare-ai/census-analysis` (Agent Skill) - Install (CLI): `npx skillmds@latest add magicare-ai/census-analysis` - Raw SKILL.md: https://api.skillmd.com/api/skills/magicare-ai/census-analysis/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - License: MIT - Author: magicare-ai (https://skillmd.com/u/magicare-ai) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/magicare-ai/census-analysis --- # Census Analysis Turn a facility's census data into a decision-ready brief. The audience is SNF administrators, admissions directors, and regional ops — often non-technical. Do the math for them, explain what the numbers mean, and always end with concrete actions, not just statistics. ## Privacy first Census data can contain PHI (resident names, MRNs, room-level detail tied to identity). Before analyzing: - If a file appears to contain resident names or identifiers, remind the user they can strip names/IDs before sharing — aggregate counts are all this analysis needs. - Treat any data provided as confidential. Never echo resident names or identifiers back in output. Report only aggregates (counts, percentages, averages). ## Step 1 — Gather inputs Accept either: 1. **A census export** (CSV/Excel — a daily or midnight census report from the EHR, e.g. PointClickCare, MatrixCare, Netsmart). Read it with available file/spreadsheet tools. Typical columns: date, unit/room/bed, payer, admission date, and sometimes anticipated discharge date. Identify the date range and whether rows are resident-days or one row per resident. Aggregate exports also work: one row per day with payer-count columns (date, Medicare, MA, Medicaid, … , total) — no resident-level detail needed. 2. **Typed numbers** — current census, licensed beds, counts by payer, admissions/discharges for the period. Always establish, asking if missing: - **Licensed beds** — the state-licensed capacity. Distinguish from **certified beds** (the Medicare/Medicaid-certified count), which can be fewer than licensed. - **Operational (available) beds** — beds actually in service. These often differ: closed wings, rooms converted to private, staffing-limited beds, isolation rooms held offline. Occupancy on operational beds is the number the team can actually manage to — lenders often use it; surveyors and CMS reference certified beds; market reports usually cite licensed. Compute licensed and operational at minimum. - **Target occupancy** (if the user has one) — needed for backfill math. If only a single day's snapshot is provided, compute point-in-time metrics and say plainly that trends need at least 2–4 weeks of daily data; offer to analyze more history if they can export it. ## Step 2 — Core metrics Compute and show the work (formulas make the brief trustworthy and reusable): **Occupancy** - Occupancy % (licensed) = census ÷ licensed beds × 100 - Occupancy % (operational) = census ÷ operational beds × 100 - Over a period, use ADC in the numerator instead of a single day's census. **Average daily census (ADC)** - ADC = total resident-days in period ÷ days in period - Compute overall ADC and ADC by payer. **Payer mix %** — each payer's ADC ÷ total ADC × 100. Standard buckets: - Medicare FFS (Part A) - Medicare Advantage (MA) - Medicaid (traditional/FFS) - Managed Medicaid - Commercial / other insurance - Private pay - (Other as present: hospice, VA, respite — keep whatever the data shows.) Map the facility's payer labels into these buckets; show the mapping if it required judgment so the user can correct it. **Skilled mix %** = skilled-level ADC ÷ total ADC × 100, where skilled typically means Medicare FFS + MA + other insurance paying for skilled care (commercial skilled, managed care skilled). Skilled mix drives revenue per patient day in most SNFs, so it gets its own headline number — a building can hold total occupancy flat while quietly trading skilled days for long-term custodial days, which is a revenue decline hidden inside a stable census. Also report **FFS vs MA composition within skilled** — skilled mix alone scores FFS→MA substitution as flat, and it isn't. Note too that census cannot see PDPM/case-mix acuity, the adjacent revenue metric — flag it as a separate analysis, not something this brief measures. **Flow** - Admissions per week, discharges per week (include deaths and hospital transfers that end the stay), and **net = admits − discharges**. Net tells you which way census is drifting even before the ADC line moves. **Average length of stay (ALOS), by payer** - Discharge-based (preferred when discharge data exists): ALOS = total days of stays that ended in period ÷ number of discharges - If only census and discharge counts exist, approximate: ALOS ≈ ADC ÷ (discharges per day) - Report by payer. Short and shrinking skilled ALOS with flat admissions means fewer skilled days; note it when it appears. Label the method used — the two formulas can differ materially. Never quote a blended (all-payer) ALOS as skilled ALOS in a mixed building — long-stay custodial residents swamp the average; per-payer discharge data is required. ## Step 3 — Trends With multi-week data: - **Week-over-week and month-over-month movement** for occupancy, ADC by payer, skilled mix, and net admissions. Show direction and size (points/percent), not just current values. - **Payer-mix shift flags.** A skilled mix decline of roughly 2+ points over a few weeks is a revenue red flag — flag it and point at the usual causes to investigate: referral acceptance rate, hospital referral volume, MA authorization denials/short auths, competitors, and discharge timing. - **Rising long-term/custodial share** while skilled falls: occupancy may look healthy while revenue quality erodes — say so explicitly. - **Volatility**: if daily census swings widely, note it (staffing and scheduling implication), and prefer weekly averages for trend claims. **Discharge pipeline.** If the data includes anticipated/planned discharge dates, build a 7-day and 14-day forward view: how many residents are expected to leave, by payer. This converts census from a rear-view metric into a forecast — the admissions team needs to know today how many beds open next week. **Backfill need** — admits per week required to hold a target census. Show the formula in the brief: ``` Required admits/week = expected discharges/week + (target ADC − current ADC) ÷ weeks to reach target ``` To simply hold current census, the second term is zero: admits must match discharges. If no anticipated-discharge data exists, use the trailing average discharges/week as the expected value and say that's the assumption. Compare required admits/week to the actual recent admit rate — the gap is the headline for the admissions team. ## Step 4 — Bed management considerations Raw "beds available" overstates true availability. When the data supports it (room/unit columns), or as a checklist otherwise, remind the user: - **Gender/room compatibility** — a semi-private bed is only available to a resident compatible with the current occupant; a building can be 90% occupied yet unable to place the next male referral. - **Isolation needs** — beds held for infection precautions or residents requiring private rooms reduce effective capacity. - **Bed holds** — beds reserved during hospital or therapeutic leave. Whether and how bed holds are paid **varies by state Medicaid policy and by managed care/MA contract** — do not assume; instruct the user to check their state rules and specific contracts before counting held beds as revenue or as available. - Distinguish **physically vacant** vs **truly sellable** beds in the brief when possible. ## Step 5 — Charting (optional) If the environment can render and share files, offer a simple trend chart — occupancy % over time and a stacked payer-mix view are the two that earn their place. Save as an image or a small self-contained HTML file. Keep it plain: dates on the x-axis, one or two series, labeled directly. If charts aren't possible, use a markdown table or a compact ASCII sparkline (e.g. `Occ%: 88 87 89 91 90 ▁▁▂▃▂`) — never skip the trend view entirely. ## Output format Produce the brief in exactly this structure: ```markdown # Census Brief — [Facility name] — [date or period] ## Headline metrics | Metric | Value | vs prior period | |---|---|---| | Census (latest) | 92 | +3 | | Occupancy % (licensed, 120 beds) | 76.7% | +2.5 pts | | Occupancy % (operational, 110 beds) | 83.6% | +2.7 pts | | ADC (period) | 90.4 | +1.8 | | Skilled mix % | 22.1% | −2.4 pts | | Admits / Discharges / Net (per wk) | 8 / 9 / −1 | net down 2 | ## Payer mix | Payer | ADC | Mix % | Trend | ALOS | |---|---|---|---|---| | Medicare FFS | ... | ... | ▲/▼/— | ... | | Medicare Advantage | ... | ... | ... | ... | | Medicaid | ... | ... | ... | ... | | Managed Medicaid | ... | ... | ... | ... | | Commercial | ... | ... | ... | ... | | Private | ... | ... | ... | ... | ## Trends - [2–4 bullets: WoW / MoM movement in occupancy, skilled mix, net flow] ## Discharge pipeline & backfill - Anticipated discharges next 7 days: N (X skilled) · next 14 days: N - Required admits/week to hold [target]% occupancy: N (formula shown above); current admit rate: N/week → gap: ±N ## Red flags - [Only real ones. e.g. "Skilled mix down 2.4 pts in 3 weeks — revenue risk."] ## Recommended actions - [Specific and assignable, e.g. "Skilled mix down X pts: review referral acceptance rate and denial reasons this week; increase hospital liaison touches at top 2 referral sources."] - [e.g. "Net flow −1/wk: admissions needs 9 admits/wk to hold 88% occupancy."] ``` Rules for the brief: - Fill every section; if data is missing for one, state what export or number would unlock it rather than leaving it blank. - Round to one decimal for percentages, whole numbers for census counts. - Red flags must be earned by the data — no boilerplate warnings. - Recommended actions must tie to a specific metric movement. - Do not state benchmark dollar figures or "industry average" rates as fact; if the user wants comparisons, direct them to their own historical data, state cost reports, or market data they have access to. ## Disclaimer This skill provides educational and operational analysis only. It is not financial, legal, reimbursement, or clinical advice. Payment rules — including bed-hold policies, Medicaid rates, and managed care terms — vary by state and contract; verify against current policies and your own agreements before acting on any figure produced here.