Should We Filter to
a Single Year?
Hospital Zeta has data from January 2023 to December 2024.
Would limiting to 2024 reveal sharper insights — or just lose information?
Two Years of Data. Do We Need Both?
2.27 million billing rows
68,441 cases
CHF 33.2M billed
Full picture, but is there noise?
1.14 million billing rows
33,805 cases
CHF 17.1M billed
Cleaner cut, but do we lose signal?
Let's Compare
2023 vs 2024, Side by Side
Seven dimensions, one conclusion
Billing Volume Is Nearly Identical
1,133,000 billing rows
34,226 distinct cases
6,138 distinct materials
86.1% of rows = CHF 0
1,141,000 billing rows
35,765 distinct cases
6,057 distinct materials
86.1% of rows = CHF 0
Verdict: stable. The CHF 1M increase (~6%) is consistent with normal healthcare cost inflation.
Patient Volume: Virtually the Same
2024: 1,136,061 rows
Change: +0.6%
2024: 31.2% of cases have a DRG code
Change: +1.6pp
Brig vs Visp: Identical Split
The Brig/Visp revenue share is 89.8% vs 89.9%. Year-over-year deviation: 0.1 percentage points. The structural dominance of Brig's oncology department is a constant, not a trend.
Same Products Dominate Both Years
| Rank | 2023 | CHF | 2024 | CHF |
|---|---|---|---|---|
| 1 | Keytruda | 2.15M | Keytruda | 2.26M |
| 2 | Darzalex | 812K | Darzalex | 993K |
| 3 | PRIVIGEN | 720K | PRIVIGEN | 761K |
| 4 | Opdivo | 500K | Perjeta | 386K |
| 5 | Ogivri | 422K | Opdivo | 380K |
The top 3 are identical in both years. Positions 4–5 shuffle slightly (Perjeta rises, Ogivri drops), but the same cancer drugs dominate throughout.
Average Billed Amount: Steady
non-zero billing row
2023
non-zero billing row
2024
change
A 5.6% increase in average unit cost is consistent with pharmaceutical price inflation. No structural break — the same pricing patterns apply in both years. A single-year view would give no baseline to judge whether prices are drifting.
What About the I/O Gap?
This is where the years get interesting
The Warehouse Gap Is Stable Across Years
| Year | Transfer Qty | Usage Qty | Avg Coefficient | CHF Gap |
|---|---|---|---|---|
| 2022 | 1,514,186 | 0 | 0.000 | -11.7M |
| 2023 | 1,277,203 | 600,532 | 0.551 | -5.0M |
| 2024 | 1,288,180 | 614,554 | 0.625 | -5.4M |
2023 and 2024 show the same pattern: about 55–63% of transfers are documented. The 2023→2024 improvement (+7.4pp) is only visible because we have both years.
2022: Ghost Movements With No Matching Data
from 2022
in 2022
in 2022
The movement file contains 303K rows from 2022, but the case and billing data only starts in January 2023. This creates an artificial I/O gap of CHF 11.7M that doesn't represent real waste — it's just an incomplete data export. This should be filtered at the probe level, not by discarding entire years.
What Would We Lose
by Filtering to 2024?
The cost of cutting a year
Five Things a Single Year Can't Tell You
- Trends vanish — The I/O coefficient improved from 0.551 to 0.625. With one year, you'd never know.
- No price baseline — Average cost rose 5.6%. Is that normal? You need a prior year to compare.
- Seasonal patterns hidden — Q4 billing dips and summer lulls need two cycles to distinguish from noise.
- Weaker probe evidence — Same material flagged in both years = structural issue. Same material flagged in one year = maybe a data entry error.
- Half the sample — 22 probes produce 223K findings. Halving the data halves each probe's confidence.
What Would We Gain From Filtering?
Can say "this is current year data" in a pitch
Removes the 2022 movement anomaly (but that should be a probe-level fix)
Lose cross-year validation (persistent vs one-off issues)
Lose the pricing baseline
Halve sample size for every probe
No probe changes its conclusion — all 22 fire the same patterns in both years
Test Results at a Glance
| Dimension | 2023 | 2024 | Difference |
|---|---|---|---|
| Billing volume | CHF 16.1M | CHF 17.1M | +6% |
| Cases | 34,636 | 33,805 | -2.4% |
| Brig CHF share | 89.8% | 89.9% | +0.1pp |
| Zero-amount rows | 86.1% | 86.1% | 0.0pp |
| Avg. billed (non-zero) | CHF 101.90 | CHF 107.64 | +5.6% |
| I/O coefficient | 0.551 | 0.625 | +13% |
| Top 3 materials | Keytruda, Darzalex, PRIVIGEN | identical | |
Every dimension tells the same story: structurally stable, no break between years.
Our Recommendation
The two-year view is strictly more informative.
Splitting adds no insight. It only removes signal.
The only fix needed is probe-level filtering of the 2022 movement anomaly —
not a wholesale year restriction that would blind us to trends.
More Data, More Confidence.
2 years · 68,441 cases · 7 comparison tests
All stable. All pointing the same way.
The strength of the analysis is the two-year window.
Thank you.