DDHA 8703 Week 9 Predictive Model Brief Example

What the Readmission Risk Score Will Do for the Transition Nurses, and What It Will Not: A Brief for the People Who Act on the Score Without Seeing the Model

Student Name

Walden University

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

Instructor Name

Month Day, Year

What the Readmission Risk Score Will Do for the Transition Nurses, and What It Will Not: A Brief for the People Who Act on the Score Without Seeing the Model

Purpose

From next month, Northgate Health's care transition nurses will receive a daily list of adult medical patients being discharged, each with a readmission risk score from 0 to 100, and will call the highest-scoring patients within two days of discharge. This brief explains what the score is, how well it performed when tested on Northgate's own patients and what it cannot do. It is written for the nurses and their managers, who will act on the score without examining the model.

Where the Score Comes From

The score comes from a logistic regression model, a standard and transparent statistical method. Its inputs are the components of the LACE index, which was derived and validated in Ontario to predict death or unplanned readmission within 30 days of discharge (van Walraven et al., 2010): length of stay, whether the admission was urgent, the patient's burden of chronic disease and the count of emergency visits over the previous half year. Northgate's analytics team added one input, the number of hospital admissions in the prior year. The model was fitted on 41,200 adult medical discharges from Northgate's four hospitals in 2022 and 2023. It was not built on surgical, obstetric or pediatric patients, and it should not be used for them.

How It Performed, in the Terms You Will Use

The model was tested on 21,000 adult medical discharges in 2024 that were not used to build it. In that test, 12% of patients were readmitted to a Northgate hospital within 30 days. The figures below come from that validation.

If the nurses call the 20% of patients with the highest scores, about 4,200 patients a year, roughly 26% of them would be expected to be readmitted, about twice the overall rate. Those calls would reach about 43% of all patients who are later readmitted. The other 57% of readmitted patients would not be on the list, because their scores were lower. Conversely, about three in four flagged patients would not have been readmitted even without a call.

A standard summary of how well the model separates readmitted from non-readmitted patients was about 0.69 in the validation, where 0.5 is chance and 1.0 is perfect. That is similar to the original index's performance in Ontario and typical of readmission models; Kansagara et al. (2011), reviewing published readmission risk models, concluded that the majority separated readmitted from non-readmitted patients only modestly.

What the Score Cannot Do

It does not identify preventable readmissions. A high score means a patient resembles patients who were readmitted, not that a call will prevent their readmission. Some high-risk patients will be readmitted whatever happens.

It does not see social needs. The model contains no information about housing, transportation, caregivers or whether a patient can afford medications, which often matter after discharge. A nurse who knows a low-scoring patient is going home alone without a ride to follow-up should not let the score overrule that knowledge.

It does not count readmissions to other hospitals. Patients readmitted outside Northgate were not counted as readmitted when the model was built, so patients who use other hospitals may receive scores that are too low.

It does not stay accurate on its own. Patient populations and practices change, and the model will be rechecked every year.

It should not be assumed fair. Obermeyer et al. (2019) found that one commercial tool, trained to forecast spending, ranked Black patients as lower risk than equally ill White patients, since less had historically been spent on their care. Northgate's model predicts readmission rather than cost, which avoids that particular problem, but its performance has not yet been checked separately for each racial and ethnic group or payer. That check will be completed before the first annual review, and the nurses will see what it finds.

How to Use the Score

Use the score to decide whom to call first, not whom to ignore. Call patients from the top of the list each day. Add any patient you are concerned about regardless of score, and record why, so that the analytics team can learn what the model misses. Do not explain the score to patients as a prediction about them.

Who to Contact

Questions about the score go to the analytics team's model owner, named on the daily list. Concerns that the score seems wrong for a type of patient should be reported the same way, since those reports will shape the annual review.

References

Kansagara, D., Englander, H., Salanitro, A., Kagen, D., Theobald, C., Freeman, M., & Kripalani, S. (2011). Risk prediction models for hospital readmission: A systematic review. JAMA, 306(15), 1688-1698. https://doi.org/10.1001/jama.2011.1515

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453. https://doi.org/10.1126/science.aax2342

van Walraven, C., Dhalla, I. A., Bell, C., Etchells, E., Stiell, I. G., Zarnke, K., Austin, P. C., & Forster, A. J. (2010). Derivation and validation of an index to predict early death or unplanned readmission after discharge from hospital to the community. CMAJ, 182(6), 551-557. https://doi.org/10.1503/cmaj.091117