← All work
Healthcare Analytics2026 / 16

Capacity & Mortality

A hospital-capacity study that asks whether occupancy is just an efficiency metric or an early signal of operational pressure tied to mortality.

PythonRegressionDiagnosticsHealthcare Ops
Source & deliverables
p<.001simple-model association
-2.87estimated slope
Rural/Urbancontext split
HTMLreproducible report
01 / Context

The question
behind the work.

Hospital occupancy is commonly treated as an efficiency measure, but its relationship with mortality is confounded by hospital type, case mix, and the difference between average utilization and true surge pressure.

My role

I defined the research question, prepared the hospital-level measures, built simple and adjusted regression models, checked diagnostics, compared rural and urban patterns, and wrote the report.

02 / Implementation

What I built.

The analysis pairs regression with hospital context and diagnostics so the operational question stays grounded in what the data can actually show.

  • Modeled risk-adjusted heart-failure mortality against capacity utilization.
  • Added hospital context and rural-versus-urban comparisons instead of stopping at a bivariate result.
  • Produced diagnostics and relationship plots to examine whether the fitted model was credible.
03 / In detail

Treating utilization as a systems signal

I built a hospital-level analysis around capacity utilization, bed size, teaching status, rural context, and risk-adjusted heart-failure mortality. The goal was to see what occupancy reveals once it is placed beside the hospital characteristics that shape how capacity is used.

The workflow starts with a simple linear relationship, then adds controls, a quadratic occupancy term, rural interaction, and diagnostic checks. That sequence makes the change in the story visible as the model becomes more realistic.

The result was more interesting than the expected story

The simple model estimated a negative occupancy slope of about -2.8731 with p below 0.001. The stronger quadratic specification kept the negative direction and found meaningful curvature, with a centered linear term near -1.749 and a quadratic term near -4.4503.

Rural hospitals remained at a higher modeled mortality level than similar urban hospitals. The interaction analysis separated that baseline difference from the shape of the occupancy relationship itself.

Making the finding useful

The report includes relationship plots, rural-versus-urban prediction curves, residual and influence diagnostics, and a full self-contained analysis. It turns the headline result into an operating question about capacity, service mix, referral patterns, staffing, and hospital capability.

That is the part I cared about most: showing how an apparently simple throughput measure can carry a deeper systems story once the model and context are built around it.

04 / Engineering judgment

The decisions
that shaped it.

  1. Kept the counterintuitive negative simple-model association visible rather than forcing the expected story.
  2. Separated statistical association from causal claims and treated occupancy as an imperfect systems signal.

Evaluation & results

The simple linear model estimated a slope near -2.8731 with p < 0.001. Additional context and diagnostics were used to show why statistical significance alone does not resolve the operational interpretation.

05 / Working outputs

See it for yourself.

Select an image to view it at full size.

06 / Artifacts

Take a closer look.

01Self-contained HTML report
Next projectRavel