NSG/543 Week 6: Data Mining, sample paper

Looking for Patterns in Two Years of Vigil Requests: What Data Mining Could Find, What It Found and Why a Pattern Is Not a Cause

[Student Name]

University of Phoenix

NSG/543: Database Management

Week 6 Assignment

[Instructor Name]

[Date]

The hospital, the program and all data are a composite written for a model paper.

Over five weeks, the program's data moved from a spreadsheet into a relational database, and queries and reports were used to answer the steering committee's questions. Those questions were ones the committee already had. This paper explores data mining, which looks for patterns no one thought to ask about.

Data Mining and Knowledge Discovery

Fayyad et al. (1996) described data mining as one step in a larger knowledge discovery process: selecting data, preprocessing it, transforming it, applying data mining methods to find patterns and interpreting and evaluating those patterns to decide whether they represent useful knowledge. The distinction matters because the patterns a method produces are not knowledge until someone has interpreted them and judged whether they are valid, new and useful. Most of the value, and most of the risk, lies in the steps before and after the mining itself.

Selection and Preparation

The data selected were all 312 vigil requests from the program's two years, including those entered from the spreadsheet during migration, linked to a limited set of record data for each patient: age group, primary diagnosis category, admission source, length of stay before the request and whether comfort care had been ordered, and when. The privacy office approved the analysis as operational quality work, with data deidentified after linking, limited to the fields needed and kept in the secure analytics environment. Preprocessing removed 19 requests with missing request times from the spreadsheet period and grouped diagnoses into seven categories.

Method 1: Clustering

Clustering groups cases that are similar across several attributes without deciding the groups in advance. Using a simple clustering method on age group, diagnosis category, admission source, days from admission to request and hours from comfort care order to request, three clusters appeared.

Cluster A, 41% of requests: older adults admitted from nursing homes with pneumonia or sepsis, requested one to three days after admission, typically within six hours of a comfort care order.

Cluster B, 33%: middle-aged adults with cancer admitted through the emergency department, requested after longer stays, often a week or more, and often many hours after comfort care was ordered.

Cluster C, 26%: adults of varied ages in intensive care after cardiac arrest or major trauma, requested on the day of admission, often late in the evening.

Method 2: Association Analysis

Association analysis looks for attributes that occur together more often than chance. Applied to the requests and their outcomes, the strongest association was between requests made more than 12 hours after a comfort care order and a first volunteer arriving after the patient's death: of 58 such requests, 21 ended before any volunteer arrived, compared with 14 of 235 requests made sooner. Cluster B accounted for most of the late requests.

Interpretation

The patterns suggest a finding the committee had not considered. Queries in Week 4 showed that late-evening requests wait longest for a volunteer, which pointed toward overnight recruitment. Data mining adds a second, earlier gap: in patients with cancer on medical units, vigils are often requested long after comfort care begins, sometimes too late. A possible explanation is that nurses caring for patients with long stays and slower declines wait for a clear sign of imminent death before requesting a vigil, while the program's volunteers need time to be scheduled.

Testing the Pattern

Before acting, the committee should test whether the pattern is real and what explains it. Three checks are planned. First, a record review of 20 Cluster B cases to see whether delayed requests reflect a clinical pattern, such as uncertain prognosis, or a workflow pattern, such as unclear responsibility for requesting. Second, conversations with oncology nurses about when they think of requesting a vigil. Third, a comparison with the next six months of data to see whether the pattern persists.

What the Pattern Cannot Show

The association between late requests and volunteers arriving too late does not prove that earlier requests would have prevented every missed vigil; some patients die faster than any schedule allows. Clusters are shaped by the attributes chosen for the analysis, and different attributes could produce different groups. And the spreadsheet period's data are less reliable than the database period's, which may affect the findings. Fayyad et al. (1996) stress evaluation as part of discovery for exactly these reasons.

Why the Database Made Mining Possible

Mining was possible only because the database stores requests, shifts and outcomes in related tables that can be linked to record data by key. The relational structure let the analysis combine a vigil with its shifts, its unit and its patient's clinical attributes without copying data into a new spreadsheet (Codd, 1970). Nursing informatics texts describe data mining as one of the ways stored clinical data can generate new knowledge when the data are well structured and their meaning is understood (McGonigle & Mastrian, 2022).

Tools and Skills

The analysis used the hospital's analytics software with its built-in clustering and association functions, run with support from a data analyst. The informatics nurse's role was to choose the attributes, prepare the data, interpret the clusters in clinical terms and decide which patterns were worth testing, which is where clinical knowledge matters most.

Privacy and Ethics

Mining data about dying patients requires care. The analysis used deidentified data, limited fields and a secure environment, and results are reported only in aggregate. The purpose, improving a program for patients, fits the reason the data were collected. Any publication beyond the hospital would require review by the institutional review board.

Communicating the Findings

The committee will receive the findings as a one-page summary that states the three clusters, the association with late requests and, in the same font size, the planned checks and the limits. Presenting a pattern as a question to test rather than a conclusion protects the program from acting on a false lead and invites oncology nurses into the inquiry rather than making them its subject.

Possible Actions

If the checks confirm the pattern, two actions follow. A prompt could appear when comfort care is ordered for a patient without family visits documented, suggesting a vigil request. And oncology units could receive a brief education session. Both would be tested for their effect on the next year's data.

Next Round of Discovery

The next round will add attributes the first analysis lacked, such as whether the patient had a documented surrogate and whether family had visited earlier in the stay, to see whether the clusters sharpen.

Conclusion

Data mining placed within the knowledge discovery process found three distinct groups of vigil requests and an association between late requests after comfort care orders and vigils that began too late, concentrated among patients with cancer on medical units. The finding is a hypothesis, not a conclusion, and will be tested before action. Used this way, data mining adds to what queries and reports show without claiming more than a pattern can support.

References

Codd, E. F. (1970). A relational model of data for large shared data banks. Communications of the ACM, 13(6), 377-387. https://doi.org/10.1145/362384.362685

Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). The KDD process for extracting useful knowledge from volumes of data. Communications of the ACM, 39(11), 27-34. https://doi.org/10.1145/240455.240464

McGonigle, D., & Mastrian, K. G. (2022). Nursing informatics and the foundation of knowledge (5th ed.). Jones & Bartlett Learning.