| Course | NURS 8210 Transforming Nursing and Healthcare Through Technology |
|---|---|
| Week | Week 9 |
| Paper type | Discussion post with two peer replies |
| Length | About 690 words, 4 pages |
| Format | Discussion post, APA 7 citations |
| School | Walden University |
| Program | Doctor of Nursing Practice (DNP) |
| Updated | September 2026 |
Free sample paper for NURS 8210 Week 9
Week 9 Discussion: Initial Post
What Was the Model Trained to Predict? A Vendor Readmission Score, a Small Pool of Transitional Care Nurses and the Case for Local Validation
My composite health system is considering a vendor model that scores every discharged adult for risk of readmission within 30 days. The system employs a small team of transitional care nurses who can follow only a fraction of discharged patients, so the proposal is simple: each morning, the patients with the highest scores go to the nurses' outreach list, and the rest receive the usual discharge call. It sounds like pure efficiency, a limited resource directed to the people who need it most. The question I want to raise is what the model was trained to predict, because that decides who the nurses will call.
A model has to be trained on a label, the outcome it learns to predict. Readmission seems like the obvious label, but vendors sometimes train on cost or utilization instead, since those data are easier to get. The choice matters. Obermeyer et al. (2019) studied a widely used algorithm that identified patients for care management programs and found that it treated spending on a patient's care as a stand-in for how much care that patient needed. Since the system had historically spent less on Black patients at any given burden of disease, a Black patient given a particular score was typically sicker than a White patient given the same score, and fewer Black patients were referred for extra help. A readmission model trained on cost, or on utilization that reflects access as much as need, could do the same thing to our outreach list.
Two more limits matter here. First, performance may not transfer. A model built on the vendor's data from other health systems may perform differently on our population, which is older and more rural than many. A systematic review found that most readmission risk models performed poorly, separating those who returned from those who did not only modestly better than chance in many cases (Kansagara et al., 2011). Second, performance drifts. Studies of clinical prediction models have shown that calibration can deteriorate over time as practice and patient populations change (Davis et al., 2017). A model validated at purchase could quietly become less accurate within a year or two.
My position: no allocation of nurses' time based on this score until it has been validated on our own patients, including its performance in subgroups by race, age, rural residence and payer, and until the label it predicts is known. After deployment, performance should be monitored on a schedule, and transitional care nurses should keep the authority to add patients the model misses, since they often know from the discharge conversation who is going home to trouble.
Question for the thread: has anyone asked a vendor what outcome their model was trained on, and what answer did you get?
Reply to Jordan
Jordan, you said the vendor has already validated the model on millions of patients, so local validation would be redundant. I would push back on that. The size of the vendor's data does not tell us how the model performs on our population, and the review by Kansagara et al. (2011) is a reminder that readmission models often perform modestly even where they were built. Local validation does not have to be elaborate: run the model silently on a few months of our discharges, then compare its scores with who was actually readmitted. If it performs well, we lose little. If it does not, we avoid directing scarce nurses to the wrong patients.
Reply to Amara
Amara, I agree with your point that nurses should be able to see why a patient scored high. I would extend it. Transparency is useful not only for trust but for catching the label problem. If nurses see that a patient's high score comes mostly from prior emergency visits, they can ask whether the model is measuring need or access. Obermeyer et al. (2019) found the bias by comparing the algorithm's scores with direct measures of illness; nurses reviewing explanations case by case could notice similar patterns on a smaller scale.
References
Davis, S. E., Lasko, T. A., Chen, G., Siew, E. D., & Matheny, M. E. (2017). Calibration drift in regression and machine learning models for acute kidney injury. Journal of the American Medical Informatics Association, 24(6), 1052-1061. https://doi.org/10.1093/jamia/ocx030
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
Reading the NURS 8210 Week 9 instructions
The Week 9 thread commonly has you judge a predictive analytics tool in your setting, including its benefits, risks and ethical implications. Expect a first post of roughly five hundred words and two replies. Describe the decision the model would drive, not only the model. Ask what outcome it was trained to predict and why that matters for who is flagged. Address whether performance transfers to your population and how it might change over time. Take a position with conditions, such as local validation and clinical override. Cite evidence on bias and model performance. Finish by asking peers what vendors told them when they asked.
How the NURS 8210 Week 9 example is put together
The opening paragraph describes the outreach list and asks the label question. The second explains labels and cites Obermeyer and colleagues on an algorithm that treated spending as need. The third names two limits, weak performance in the review by Kansagara and colleagues and drift shown by Davis and colleagues. A highlighted fourth paragraph states the position: validation on local patients by race, age, rurality and payer, scheduled monitoring and nurse authority to add patients. A closing question asks about vendor answers. The reply to Jordan proposes running the model silently on a few months of discharges. The reply to Amara links transparency to spotting the label problem.
NURS 8210 Week 9 rubric: what earns full marks
Predictive analytics posts are typically graded on understanding of how models work, attention to bias and equity, a reasoned position and peer engagement. Understanding earns credit when the post explains labels, validation and drift in plain terms. Bias credit rewards specific mechanisms supported by research rather than general concern. The position is judged on whether it is conditional and practical. Replies earn credit when they answer objections with methods a real organization could use. Credible sources, correct references and punctual posting make up the remainder. Faculty also value posts that keep a role for nursing judgment alongside the model's output. Faculty often give extra credit to a post that suggests a validation method small enough for a single health system to run.
NURS 8210 Week 9 help: mistakes that cost marks
Posts on predictive tools lose marks when they praise or condemn algorithms in general without a specific decision. Another common loss is citing bias research without explaining how it applies to the tool at hand. Some posts accept vendor validation at face value. Others propose validation methods too elaborate for any organization to run. Describe the decision, ask about the label, address transfer and drift, take a conditional position and suggest practical validation. The desk can draft a post about a predictive tool used or proposed in your setting. Ask your informatics team what outcome your organization's models predict before writing, since the answer is often surprising. If nobody knows, that gap is itself worth raising in the post.
Get a NURS 8210 Week 9 example written to your instructions
Emerging-technology prompts may name predictive analytics, generative tools or wearables; paste the prompt as written, with its rubric, so the thread matches. The thread comes back in 24-48h, post and replies together, the first free, with every limitation tied to a named source rather than a general caution.
More NURS 8210 papers and DNP samples
- NURS 8210 Week 1: Technology Transformation Post
- NURS 8210 Week 2: Informatics Maturity Post
- NURS 8210 Week 3: Technology Outcome Rationale
- NURS 8210 Week 4: Decision Support Governance Paper
- NURS 8210 Week 5: Socio-Technical Analysis Paper
- NURS 8210 Week 6: Virtual Care Safety Debate Post
- NURS 8210 Week 7: Data Stewardship Analysis Paper
- NURS 8210 Week 8: Technology Business Case
- NURS 8210 Week 10: Informatics Evaluation Plan
- NURS 8210 Week 11: Technology Leadership Synthesis
- NURS 8114 Week 9: Operationalizing Concepts Post
- NURS 8100 Week 9: Legislative strategy thread
- NURS 8003 Week 9: Development plan
- NURS 8002 Week 9: DNP change agent post
NURS 8210 Week 9 questions, answered
Does the post argue against using predictive models?
No. It argues for adoption under conditions, local validation including subgroup performance, retained clinical override and named monitoring ownership. Rejecting prediction outright would ignore the real limits of unaided judgment in allocating scarce nursing time. The doctoral contribution is specifying the conditions under which the organization could trust a score with a resource decision.
Why cite Obermeyer and colleagues specifically?
Because their study shows the precise mechanism the post is worried about: a widely used algorithm predicted health costs, and because less money had historically been spent on Black patients with the same needs, the scores understated their illness. That concrete mechanism is stronger evidence than general statements about bias, and it maps directly onto a readmission model allocating nursing resources.
Which vendor's model is being discussed?
None in particular. The model stands in for the commercial readmission tools on the market generally, and the health system around it is invented. Naming a product would invite claims about its performance that a discussion post could not support. The limits argued, label choice, transfer and drift, apply to any clinical prediction model, whoever builds it.
Where can I find a free NURS 8210 Week 9 sample paper?
The post on a vendor readmission model is free to read here together with its two replies and three sources. A post about a predictive tool in your own setting can be requested.
What conditions does the NURS 8210 Week 9 post set before using the score?
Knowing the label the model predicts, validating it on local patients with subgroup results, monitoring it on a schedule and letting nurses add patients it misses.