Standardized Mortality Ratio (SMR) & Hospitalization Rate
A Standardized Mortality Ratio (SMR) compares the deaths your unit actually had (observed) against the number a comparable, case-mix-adjusted reference population would be expected to have (expected) over the same period: SMR = Observed ÷ Expected. An SMR of 1.0 means your unit matches expectation; below 1.0 is better than expected; above 1.0 is worse. The same logic, applied to hospital admissions instead of deaths, gives a Standardized Hospitalization Ratio (SHR).2
Patient-months at risk ≈ average prevalent patient count × number of months in the period (e.g., 100 patients tracked for 12 months = 1,200 patient-months). Both panels below use this same figure to convert counts into a per-patient-year rate.
Where "expected deaths" comes from: ideally your registry's case-mix-adjusted expected count (age, sex, diabetes status, vintage) — in the Philippines this may come from PRDR/REDCOP if your registry publishes it. Lacking that, some units approximate it as patient-years at risk × a reference crude mortality rate for a comparable population; treat that approximation as cruder than a true case-mix-adjusted expected count, and say so when you report it.
The field manual's own dashboard (Section 6.7) sets no single hard numeric target for this metric — "trend down" is the standard — so this tool reports the computed rate plainly and, if you supply a facility target or last period's rate, shows whether you're trending toward or away from it. Supplying "expected admissions" (from a registry that publishes case-mix-adjusted expected counts) additionally yields a statistically grounded SHR.
Reading the Output
Suppose a 100-patient unit tracked for a full year (1,200 patient-months) recorded 8 deaths against a registry-supplied expected count of 6.5, and 170 hospital admissions against a facility target of 1.5 admissions per patient-year. Loading this example (button above) gives:
| Metric | Result | Read |
|---|---|---|
| Crude mortality rate | 8 deaths / 100 patient-years | The raw rate, useful for trending even without an expected count |
| SMR | 1.23 (95% CI 0.53–2.43) | Point estimate is in the "red" band (>1.2), but the wide confidence interval crosses 1.0 — with only 8 events, this is not yet statistically distinguishable from expected. Keep watching; don't over-react to one period. |
| Hospitalization rate | 1.7 admissions / patient-year | Above the 1.5 facility target — flagged for the CQI meeting, but not (by itself) a sentinel finding |
This is exactly the discipline Section 5.2 of the field manual teaches for run charts, applied to a single-period ratio: a point estimate alone can mislead; the confidence interval (or, over time, a run chart of the ratio) tells you whether to act now or keep collecting data.
How the Ratios Are Computed
SMR = Observed deaths ÷ Expected deaths; SHR = Observed admissions ÷ Expected admissions — the same standardized-ratio logic CMS and USRDS use to compare US dialysis facility performance after adjusting for patient case mix.2 The crude rate (deaths, or admissions, per 100 patient-years / per patient-year) needs no expected count and is always shown. When an expected count is supplied, the tool adds an exact 95% confidence interval using the chi-square–Poisson link described by Ulm (1990)1 — the same family of method used for standardized infection ratios in healthcare epidemiology. A ratio whose confidence interval excludes 1.0 is statistically distinguishable from the reference population; one whose interval includes 1.0 is not, regardless of how far the point estimate itself sits from 1.0 — this matters most with small event counts, exactly where a raw ratio is most likely to mislead. Admissions-per-patient-year is the standard unit for reporting dialysis hospitalization burden in the literature.4 Every target discussed here is explained in the QAPI vs. CQI field manual, Section 6.7, and this tool is the outcomes-domain companion to the QAPI Scorecard and the Run Chart & SPC Generator (plot either ratio's month-to-month history there to test whether a change is signal or noise).
