A $1.4 Million Courtyard and a Single Retrospective Study
Your team is designing a 30-bed cancer care pavilion and must decide whether to spend $1.4 million on an interior courtyard visible from 60% of patient rooms. A team member found a retrospective study of 214 patients at one academic hospital reporting that patients whose rooms faced a garden had a mean length of stay 1.2 days shorter (p = 0.03) than patients facing a parking structure. The study covered four years and adjusted for age and diagnosis. Your pavilion serves a population that is 40% outpatient infusion; inpatient mean LOS is 3.4 days; and rooms will be assigned by bed control on the basis of availability at the time of admission, not by view.
1. Critically appraise the study for your decision. Identify the specific threats to validity that matter most here.
Read the worked answer
Design: retrospective, single site, observational, no randomization. The comparison is view exposure as it happened, not as assigned.
Threats that actually bite:
- Confounding by room location. Rooms facing a garden are rarely randomly distributed; they may be on quieter floors, closer to nursing stations, or reserved for a particular payer or acuity. Adjusting for age and diagnosis does not touch this.
- Selection into rooms. If bed control steers certain patients to garden rooms on busy days, the exposure correlates with census and staffing - both drivers of LOS.
- Co-intervention and secular trend over four years.
- Statistical vs. practical significance. p = 0.03 in a 214-patient sample is fragile; a 1.2-day effect on a typical 4–6 day stay is a ~20–25% reduction, which is implausibly large for a view alone and suggests residual confounding or a small-sample exaggeration.
- Multiple comparisons: four years of outcomes usually yields many tested associations; one at p = 0.03 is unsurprising.
Also note the absence of a dose-response or a plausible biomarker/mechanism measure (stress, analgesia use, sleep).
2. Assess transferability to your project: population, setting, and the mechanics of how the exposure would occur in your building.
Read the worked answer
Transferability is weaker than the p-value suggests.
- Population mismatch. Your census is heavily outpatient infusion; the inpatient subgroup is likely a different acuity mix from a general academic cohort. A view effect on LOS requires inpatient exposure over days.
- Outcome headroom. Your mean LOS is 3.4 days. A 1.2-day reduction would be a third of the stay - no mechanism (stress reduction, better sleep) plausibly produces that. The transferable expectation is far smaller.
- Exposure mechanics. In the study, view was a property of the room. In your pavilion, bed control assigns rooms by availability, so exposure will be partial, uncontrolled, and diluted - and 40% of rooms will have no courtyard view, which is itself an equity and staff-morale issue worth naming.
- Comparator. Your alternative is not a parking structure; it is a well-designed room with a window to sky and landscape elsewhere on the site.
Conclusion: the study supports a plausible direction, not a magnitude, and cannot by itself carry $1.4 million.
3. What recommendation would you make under this uncertainty, and what evidence would change your position?
Read the worked answer
Recommend a decision that buys information as well as square footage.
- Reframe the question: not 'courtyard yes/no' but 'which configuration delivers daylight, nature access, and respite at the best cost per unit of benefit?' Compare the courtyard against distributed gardens, terrace access, and view corridors - often a cheaper route to the same mechanism.
- If the courtyard proceeds, fund it primarily on grounds that are well supported (staff respite, family coping, wayfinding, daylight) rather than on a contested LOS claim, and say so in the decision record.
- Build the evaluation in: bed-control-neutral assignment is impossible, but you can pre-specify view exposure per room, record it at admission, and analyze with LOS and analgesia use as outcomes, adjusted for acuity and floor.
- What would change my position: a systematic review or multi-site study with a smaller but consistent effect; a local natural experiment; or a pilot wing where random or quasi-random assignment is feasible.
Spending $1.4M on an unverified single study is a faith-based decision wearing an evidence-based label.