C797 Task 1 Clinical Data Analysis Example

What Six Months of Structured Suicide Risk Screening Data Show: A Descriptive Analysis, a Data Quality Review and Three Decisions for a Rural Emergency Department

Student Name

Leavitt School of Health, Western Governors University

C797: Data Science and Analytics, Task 1

Course Instructor

Month Day, Year

What Six Months of Structured Suicide Risk Screening Data Show: A Descriptive Analysis, a Data Quality Review and Three Decisions for a Rural Emergency Department

Purpose and Data Source

Half a year has passed since a composite rural hospital with 110 beds retired its scanned paper screening form. Emergency department (ED) nurses now complete the C-SSRS, the Columbia suicide screener, as discrete fields in the electronic health record (EHR) for all patients 12 years and older. The informatics nurse specialist has been asked to analyze the first six months of data and advise the ED nurse manager and the behavioral health liaison. The data were extracted from the reporting database built for the project, which stores each encounter, each screening with its individual answers and calculated risk level, and each intervention with its order and start times. The extract was de-identified before analysis. The analysis asks four questions: how completely eligible patients were screened, how results were distributed, how quickly high-risk patients received one-to-one observation, and whether the data are trustworthy enough to support those answers.

Data Quality Before Analysis

Data reused from an EHR must be checked before conclusions are drawn. Weiskopf and Weng (2013) identified five dimensions commonly used to assess EHR data quality: completeness, correctness, concordance, plausibility and currency. Each was checked against this extract.

Completeness: of 5,218 screenings, 5,180 (99.3%) had an answer recorded for every displayed question; the 38 incomplete records all carried a deferral reason, as the design requires. Correctness: a sample of 50 screenings was compared with the nurse's narrative note, and the risk level matched in 49; the mismatch was a nurse who documented a patient's retraction of an answer in the note but not in the screener. Concordance: screenings with a high risk level were compared with orders, and 7 of 118 had no observation order at all, which is a clinical gap rather than a data error and is discussed below. Plausibility: 11 observation start times preceded the screening time by a few minutes, all on patients already under observation for another reason; these were excluded from the timing analysis. Currency: data were available in the reporting database within 24 hours, which is adequate for monthly review but not for real-time monitoring.

Screening Completeness

The ED recorded 5,902 encounters by patients aged 12 and older over six months. Of these, 5,218 had a completed or deferred screening, a screening rate of 88.4%. The rate rose from 81.4% in the first month to 92.6% in the sixth. By shift, the rate was 91.8% on days, 89.5% on evenings and 81.2% on nights. Of the 684 unscreened encounters, 402 (58.8%) were patients who left before being seen or were transferred out within an hour of arrival, groups the triage workflow may never reach. The remaining 282 represent missed screens that the design should have prevented.

For comparison, eight EDs that adopted and then refined universal screening reached documented screening in 84% of visits, up from a starting point of 26% (Boudreaux et al., 2016). This ED's rate compares well, but the gap on nights is large enough to act on.

Distribution of Results

Of the 5,180 completed screenings, 4,872 (94.1%) were negative, 190 (3.7%) were low or moderate risk and 118 (2.3%) were high risk. Among the 420 screened patients whose chief complaint was psychiatric, 132 (31.4%) screened positive at any level, compared with 176 of 4,760 (3.7%) among patients presenting for other reasons. That difference is expected, but the second group matters: 176 positive screens came from patients who presented with problems such as abdominal pain, headache or injury and would not have been screened under the old policy. Screening advocates make exactly this point, that people at risk often come to medical settings for reasons unrelated to mental health and are missed unless everyone is asked (Horowitz et al., 2020).

Time to Observation for High-Risk Patients

Of 118 high-risk screenings, 111 had an observation order, and 100 had a valid start time after the plausibility exclusions. The median time from screening to the start of one-to-one observation was 14 minutes (interquartile range 8 to 26), compared with a median of 34 minutes observed under the paper process. Nine patients (9.0%) waited more than 45 minutes; seven of the nine were on the night shift, when one security officer covers both observation and the building. The seven high-risk screenings without any observation order were reviewed individually: four patients were already in a locked behavioral health room with a sitter, and three were genuine omissions.

Interpretation and Limitations

The data show a process that has improved and is still improving, with two weaknesses concentrated on the night shift: lower screening completeness and longer waits for observation. A run chart of monthly screening rates shows six consecutive rising points, which suggests a real trend rather than random variation, though six months is a short series and the rule should be applied with caution (Provost & Murray, 2011). The analysis is descriptive: it cannot show whether screening prevented any self-harm, and the number of high-risk patients is small enough that a few cases change the timing results noticeably.

Recommendations

First, add a night shift reminder: an EHR task to the primary nurse if a patient 12 or older has been in a room for 60 minutes without a screen. Second, work with security and nursing leadership on a night shift observation plan, such as a trained patient care technician available to the ED on nights, since the longest waits cluster there. Third, make the observation order set fire automatically for provider signature on every high-risk screen, and add the three omissions to the monthly case review. The analysis should be repeated quarterly, with the data quality checks run first each time. Two further questions deserve a larger data set than six months provides: whether screening rates differ by patient language, since interpreter use is recorded, and whether patients with positive screens return to the ED more often within 90 days. Both can be answered from the same reporting database once a full year of data exists.

References

Boudreaux, E. D., Camargo, C. A., Arias, S. A., Sullivan, A. F., Allen, M. H., Goldstein, A. B., Manton, A. P., Espinola, J. A., & Miller, I. W. (2016). Improving suicide risk screening and detection in the emergency department. American Journal of Preventive Medicine, 50(4), 445-453. https://doi.org/10.1016/j.amepre.2015.09.029

Horowitz, L. M., Roaten, K., Pao, M., & Bridge, J. A. (2020). Suicide prevention in medical settings: The case for universal screening. General Hospital Psychiatry, 63, 7-8. https://doi.org/10.1016/j.genhosppsych.2018.11.009

Provost, L. P., & Murray, S. K. (2011). The health care data guide: Learning from data for improvement. Jossey-Bass.

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