DDHA 8703 Week 7 Adjusted Comparison Example

A Day and a Half Longer on the Medical Unit, and Why That Is Not the Finding: Comparing Length of Stay on Two Units Before and After Accounting for Who Arrives

Student Name

Walden University

DDHA 8703: Advanced Health Analytics and Data-Driven Decision Making

Instructor Name

Month Day, Year

A Day and a Half Longer on the Medical Unit, and Why That Is Not the Finding: Comparing Length of Stay on Two Units Before and After Accounting for Who Arrives

The Raw Comparison

At Northgate Health's flagship hospital, a composite 420-bed teaching facility, leaders asked why patients on 4 East, a general medical unit, stay longer than patients on 5 West, a general surgical unit. In fiscal year 2025, the mean length of stay was 5.6 days on 4 East and 4.1 days on 5 West, a difference of 1.5 days. The medians were 4.2 and 3.0 days. Taken at face value, the comparison suggests the medical unit is less efficient. It cannot be read that way, because the two units serve different patients, and length of stay depends heavily on who the patient is and why they came. The rest of this paper explains what the comparison looks like once that is accounted for.

Who Arrives on Each Unit

The medical unit's patients are older, carry more chronic conditions and almost all arrive unplanned. One in seven comes from another hospital, often after a stay elsewhere. The surgical unit's patients include many elective admissions for planned procedures, whose length of stay is scheduled and short. More than twice as many medical patients go to a skilled nursing facility, which requires placement and often adds days.

Table 1

Patient Profile of Each Unit, Fiscal Year 2025

The comorbidity score is the weighted Elixhauser point score; the profile uses the hospital's discharge data for all adult discharges from each unit.

How the Adjustment Works, in Words

The adjustment asks a simple question: given the kinds of patients each unit receives, how long would we expect them to stay, and how does each unit's actual stay compare with that expectation? To estimate the expected stay, the analysis used all 31,400 adult medical and surgical discharges at Northgate's four hospitals in the same year. It modeled length of stay as a function of age, comorbidity, admission source, whether the patient was transferred in and the patient's diagnosis-related group, which captures the main reason for admission. That model was then applied to each unit's own patients to produce an expected length of stay for each patient, and the expected values were averaged. Comparing actual with expected gives an observed-to-expected ratio for each unit. A ratio of 1.0 means the unit's patients stayed as long as similar patients across the system.

Comorbidity was measured with the Elixhauser comorbidity measures, which Elixhauser et al. (1998) developed for use with administrative data, scored with a published point system derived from those measures (van Walraven et al., 2009). Discharge to a skilled nursing facility was deliberately not included in the model, because it is partly a result of what happens during the stay and partly a cause of delay, and adjusting for it would hide some of the difference the hospital is trying to understand.

Unadjusted and Adjusted Results

Most of the raw difference reflects the patients. Of the 1.5-day gap, about 1.1 days is what the model would expect from differences in age, comorbidity, admission source, transfer status and diagnosis. The remaining difference is small: the medical unit's patients stayed about 6% longer than expected, while the surgical unit's stayed about 2% less than expected. In days, that residual is about 0.4 of a day per patient.

Table 2

Length of Stay Before and After Adjustment

Expected values come from the system-wide model described above; both raw and adjusted figures are shown so that neither is hidden.

What Survives Adjustment and What It Might Mean

A residual difference of about 0.4 days survives. It may reflect something about how the medical unit works, such as the timing of discharge decisions or delays in arranging post-acute care. It may also reflect patient differences the model does not capture, especially illness severity within a diagnosis group, which administrative data measure poorly. Iezzoni (1997), discussing quality assessment from administrative data, warns that risk adjustment based on billing data cannot fully capture clinical severity, so residual differences should be treated as signals to investigate rather than as verdicts. The higher share of medical patients discharged to skilled nursing facilities, which the model left out on purpose, is the most obvious place to look.

A Sensitivity Check

Because transferred-in patients may differ from others in ways the model does not see, the adjustment was repeated without them. The medical unit's observed-to-expected ratio fell slightly, from 1.06 to 1.04, and the surgical unit's was unchanged. The residual difference shrank to about 0.3 days per patient. Transfers therefore explain a little of the residual, but most of it remains, which supports looking elsewhere, starting with discharge planning, for the rest.

Limits

The model is only as good as its variables, and the most important missing one is severity at admission. Transfers' length of stay is counted from arrival at this hospital, which understates their total time in the hospital but is appropriate for comparing units. The analysis covers one year. And an observed-to-expected ratio compares each unit with the system average, not with best practice; both units could be slower or faster than they should be.

Conclusion

The medical unit's patients stay about a day and a half longer than the surgical unit's, but most of that reflects who the patients are. After adjustment, a gap of about 0.4 days per patient remains. That residual is worth examining, starting with discharges to skilled nursing facilities, and it is far smaller than the raw comparison suggested.

References

Elixhauser, A., Steiner, C., Harris, D. R., & Coffey, R. M. (1998). Comorbidity measures for use with administrative data. Medical Care, 36(1), 8-27. https://doi.org/10.1097/00005650-199801000-00004

Iezzoni, L. I. (1997). Assessing quality using administrative data. Annals of Internal Medicine, 127(8, Pt. 2), 666-674. https://doi.org/10.7326/0003-4819-127-8_Part_2-199710151-00048

van Walraven, C., Austin, P. C., Jennings, A., Quan, H., & Forster, A. J. (2009). A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Medical Care, 47(6), 626-633. https://doi.org/10.1097/MLR.0b013e31819432e5