NSG/541 Week 2: Data Quality Assessment, sample paper

Four Minutes Fast and Fourteen Percent Canceled at the Desk: A Data Quality Assessment of Nurse Call, Record and Event Report Data Before Any Analysis

[Student Name]

University of Phoenix

NSG/541: Data Analysis and Management

Week 2 Assignment

[Instructor Name]

[Date]

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

The Week 1 map identified four sources for the unit's question about slow responses and falls: nurse call logs, bed assignment history from the electronic health record, fall reports and staffing data. Before any analysis, the data must be tested. This paper assesses a 90-day sample.

The Framework

Weiskopf and Weng (2013) reviewed how researchers assess the quality of electronic health record data and identified five dimensions: completeness, whether the data are present; correctness, whether they are true; concordance, whether sources agree; plausibility, whether values make sense; and currency, whether data reflect the time they describe. They also identified methods for checking each, including validity checks, comparison between sources and review of logs. I used this framework to organize the assessment.

The Sample

Facilities provided a 90-day extract of nurse call logs for the unit: 21,864 call records. Clinical informatics provided bed assignment history for the same period, with 1,047 patient stays, and the patient safety office provided the eleven fall reports. The unit's census recorded 2,591 patient days.

Completeness

Every call record had a room number, call type and placement time. However, 2,012 records, or 9.2%, had no bed identifier because the room is semi-private and the call came from the shared bathroom, which is wired to the room rather than to a bed. For those calls, the patient cannot be identified directly. Cancellation location was present for all records. Fall reports were complete for date, location and activity; three of eleven gave the time only as "around 0400" or similar.

Correctness

Correctness was tested by comparing a sample with direct observation. Across three shifts, two days and a night, an observer recorded the time staff entered rooms after 60 calls. For calls canceled at the bedside, the logged cancellation time was within 30 seconds of the observed entry in 55 of 57 cases. For calls canceled at the nurse station, logged cancellation did not correspond to anyone entering the room in 9 of 13 cases; staff canceled the call from the desk and either went later or not at all.

The Largest Problem: What Cancellation Means

Across the extract, 3,061 calls, 14.0%, were canceled at the nurse station rather than the bedside. For these calls, the time between placement and cancellation measures how long the call rang before someone silenced it, not how long the patient waited for help. Treating desk cancellations as responses would make the unit look faster than it is, and would hide the calls most likely to leave a patient waiting. Decision: calls canceled at the desk will be analyzed separately and will not be counted in response time. Their frequency will be reported as its own measure.

Concordance

The key concordance question is whether the nurse call system and the electronic health record agree on time. I compared the timestamps of 20 code calls, which are logged in the nurse call system and documented in the record by the responding team to the minute. The nurse call server's clock ran 4 minutes and 10 seconds ahead of the record's clock on average, with little variation. Decision: subtract four minutes and ten seconds from nurse call times before linking, and ask facilities to synchronize the server with the hospital's network time source.

A second concordance check compared bed assignments in the record with call origins: in 97% of calls from a bed, the bed had an assigned patient at that time. The other 3% came from empty beds, most often while a bed was being cleaned, and will be excluded.

Plausibility

Validity checks looked for implausible values. There were 1,318 calls, 6.0%, placed within 60 seconds of a previous call from the same room and canceled together, which usually reflects a patient pressing the button repeatedly. Decision: merge calls from the same bed within 60 seconds into one call event. There were 214 calls lasting more than 60 minutes before cancellation; review of a sample showed most were forgotten staff-assist calls. Decision: exclude staff-assist calls from patient response time and cap patient calls at 60 minutes, reporting how many were capped.

Currency

Nurse call data are written in real time and are current. Fall reports are entered after the event, typically within two hours, and the time of the fall is from recollection. Decision: use fall time as reported, and in analysis treat call-fall links within a window rather than to the exact minute.

Why Quality Must Be Checked Before Analysis

The unit's question is sensitive: the answer may suggest that staff respond too slowly. If the data are wrong, the unit could blame staff for delays that did not happen or miss delays that did. Published studies relating response time to falls, such as the four-hospital analysis by Tzeng et al. (2012), depended on how their systems recorded responses, and a unit that copies their conclusions without checking its own data may be measuring something different. Checking quality first is also required by the question itself: elimination-related falls were half of all falls in one prospective hospital study (Hitcho et al., 2004), so bathroom calls must be identified correctly, and the completeness check shows they cannot always be linked to a patient.

Summary of Decisions

Desk cancellations are separated from response time. Clock offset is corrected by four minutes and ten seconds. Shared-bathroom calls without a bed identifier are linked to the room and, where the room has one patient, to that patient; where two patients share it, the call is kept for unit-level measures but not patient-level links. Duplicate calls within 60 seconds are merged. Staff-assist calls are excluded from patient response measures. Calls over 60 minutes are capped. Approximate fall times are used with a time window.

Who Reviewed the Assessment

The data analyst reviewed each check and decision, and the unit manager confirmed the observation findings about desk cancellation. Recording who reviewed each step makes the assessment easier to defend.

Limitations the Analysis Will Carry

Even after these steps, the analysis will not know why most patients called, cannot identify the patient for some shared-bathroom calls and depends on approximate fall times. These limits will be stated with every result.

Recommendations to Improve Data at the Source

Three changes would improve future data: synchronize the nurse call server's clock; configure the system so that bathroom calls in semi-private rooms record the bed that last called, as the vendor says is possible; and educate staff that desk cancellation should be used only when the patient's need has been met by phone, with a documented reason.

Conclusion

The data quality assessment found data that are usable but not ready. The most serious problem is that 14% of calls were canceled at the desk, where cancellation does not mean help arrived. Correcting the clock offset, merging duplicates and separating desk cancellations make the analysis possible, and recording each decision lets readers judge the results. Week 3 will define the measures.

References

Hitcho, E. B., Krauss, M. J., Birge, S., Dunagan, W. C., Fischer, I., Johnson, S., Nast, P. A., Costantinou, E., & Fraser, V. J. (2004). Characteristics and circumstances of falls in a hospital setting: A prospective analysis. Journal of General Internal Medicine, 19(7), 732-739. https://doi.org/10.1111/j.1525-1497.2004.30387.x

Tzeng, H.-M., Titler, M. G., Ronis, D. L., & Yin, C.-Y. (2012). The contribution of staff call light response time to fall and injurious fall rates: An exploratory study in four US hospitals using archived hospital data. BMC Health Services Research, 12, Article 84. https://doi.org/10.1186/1472-6963-12-84

Weiskopf, N. G., & Weng, C. (2013). Methods and dimensions of electronic health record data quality assessment: Enabling reuse for clinical research. Journal of the American Medical Informatics Association, 20(1), 144-151. https://doi.org/10.1136/amiajnl-2011-000681