BACKGROUND/RATIONALE:
Hospitals are developing plans for handling the surge of patients seeking care for COVID-19 and corresponding strains on health care resources, including staff, personal protective equipment (PPE), and respiratory support equipment. This project will address the HSR&D priority area of staffing: new approaches needed to respond to the outbreak, efficient ways of expanding or shifting staff capacity, and models for dealing with staff shortages due to illness/quarantine. Difficulty in determining how many staff members are in direct patient care roles at the hospital level and at the inpatient unit level has been an ongoing challenge for both research and operations applications. As part of VA-funded work (HSR&D IIR 15-438) examining how inpatient resources and burden affect outcomes, our team has developed methods for assessing nurse staffing and workload using staff activity data in the electronic health record (EHR), allowing for identification of staff (e.g., registered nurses [RNs]) in direct patient care roles and also where in the hospital these staff are located at particular times. The EHR data are consistent across all hospitals in the VA system and collected as part of routine care, making comparisons across the entire system possible.

OBJECTIVE(S):
We aim to adapt our methods developed using historical data to real-time or near real-time data and then to work with operational partners to make staffing monitoring tools available for national, network, hospital, and/or unit leaders.
Aim 1: Adapt methods for assessing inpatient staffing to real-time or near-real-time data and provide reports to operational leaders
Aim 2: Obtain input from operational leaders on their preferences for staffing report features and incorporate prioritized and feasible items into staffing reports

METHODS:
This project utilized electronic health record (EHR) data from the VA Corporate Data Warehouse for all VHA acute-care inpatient units, as well as the Nurse Unit Mapping Application (to link bed locations with nursing units) and Bed Management Solutions data (to identify beds and units designated for COVID-19 patients). Provider tables were used to match staff IDs to staff roles.
Aim 1 (general)
We mapped staff IDs to staff roles using provider tables. A research team member categorized each staff role into one of five staff role categories: registered nurse (RN), licensed practical/vocational nurse (LPN/LVN), nursing assistant (NA), Other-nursing, and Other-non-nursing. These role categories were used to filter BCMA data by role in Aim 1a/1c and to define the staff members (i.e., RNs) for whom data would be pulled for Aim 1b.
We downloaded daily reports of Bed Management Solutions (BMS) data to identify beds and units designated for COVID-19 patients. Since BMS did not provide retrospective reports and updates in real-time, we used Power Pivot in Microsoft Excel to link our local network to BMS and manually refreshed this data every weekday around 12:00pm CT. We also pulled data for some weekends and holidays when such days were within one day of when a team member was available to pull the data. We compiled the downloaded reports into a database to be processed and integrated into the dashboard we developed for Aim 1c.
1a: Use bar code medication administration (BCMA) data to estimate staffing and workload.
1a.1 Identified the peak medication pass (medpass) time for each acute-care unit using two different methods: 1) Across days: Computed the frequency of medications scheduled at each time-of-day. The time-of-day with the highest frequency of scheduled medications was flagged as the peak medpass time; 2) Within days: Computed the frequency of medications scheduled at each time-of-day for every day of the year. Computed the number of days a given time-of-day had the highest within-day frequency. Flagged the time-of-day with the highest number of days on which it was the most-frequent scheduled time as the peak medpass time.
1a.2 Defined the peak medpass window for each unit as the two-hour period around the unit-specific peak medpass time.
1a.3 Defined basic features (number of patients, number of medications, medpass duration) of each staff member's peak-time medpass on a given day. For each nurse, all peak-time scheduled medications administered within the peak medpass window (+/- 120 minutes) were included. For each staff member, computed the number of patients to whom medications were administered and the number of medications administered within the peak medpass window on each day. We also determined each staff member's medpass duration, calculated as the total time from the first peak-time scheduled medication to the last peak-time scheduled medication scanned/administered.
1a.4 Established unit and facility level peak-time medpass staffing and workload. The number of unique patients receiving peak-time medpass medications and the number of unique staff administering peak-time medpass meds was computed at the unit and the facility levels, along with the mean number of patients and medications per staff.
1b: Estimated direct-care RN hours worked. We adapted a previously developed method for estimating the number of direct-care hours worked per inpatient RN based on activity documented in the electronic health record (BCMA, vital signs, notes). We estimated direct-care hours for each RN by adding up all 2-hour blocks in which the RN has at least one of the indicated activities in the health record and bridging any 2-hour blocks in which the RN has activity in both the preceding and following 2-hour block.
1c: Prepared reports for monitoring staffing across all VA acute-care units. Data for reports were aggregated at the unit and facility level and show trends in daily peak-medication-time nurse staffing and workload (Aim 1a), daily overall RN staffing (Aim 1b), and bed occupancy (Aim 1-general).
Aim 2
We virtually convened and/or corresponded on at least a bi-monthly basis with operational partner representatives to determine their needs for rapid development of predictive models related to inpatient staffing.

FINDINGS/RESULTS:
Aim 1 (general)
We integrated Bed Management Solutions data into the dashboard we developed for Aim 1c. Please see Aim 1c results for more details.
Aim 1a
For year 2020, BCMA data involved a total of 112 facilities and 455 acute-care nursing units. Peak medpass times were identified for each nursing unit using barcode scanning data using two methods (described in Methods section). Peak-times as determined by the two methods were in agreement for most units (99%). Units in which there was a disagreement had fewer days on which data were available. The two most-common peak times for scheduled medications across VA acute-care units were 9 a.m. (n=364 units) and 10 a.m. (n=134 units). On average, 37% of scheduled medications in a given unit were scheduled for the peak time of the unit, with variation across units. At the unit level, the second most frequent time for scheduled medications occurred most commonly 12 hours from the peak time (n=458 units). On average, 21% of scheduled medications were scheduled for the second most frequent scheduled time, with variation across units.
A total of 23,559 distinct staff members administered peak-time scheduled medications to 189,458 distinct patients. Across all days, a total of 2,762,567 patient stops were made resulting in 14,553,506 peak-time medication scans.
There was significant variability across facilities, units, and days in the number of staff, number of patients, and number of patients per staff. Preliminary data suggest that the mean numbers of patients per unit (10.45 [SD 6.60] in CY20; 12.20 [SD 7.39] in FY15), staff per unit (4.82 [SD 2.09] in CY20; 5.02 [SD 2.10] in FY15), patients per staff (2.21 [SD 0.92] in CY20; 2.62 [SD 1.17] in FY15), and medications per staff (14.78 [SD 6.86] in CY20; 15.11 [SD 8.09] in FY15) during the peak-time medication pass in CY2020 were lower than observed in the team's analysis using FY15 data across all VA acute-care units. Medication pass duration across the two datasets was similar but of greater length in the CY2020 data (59.96 minutes [SD 12.50] in CY20; 59.36 minutes [SD 14.80] in FY15). We plan additional comparisons with more recent pre-pandemic data.
We identified COVID surge periods for many VA units, including transition points for turning regular units to COVID units, based on observably significant staffing changes in our data and in peak medpass metrics.
Aim 1b
The total number of estimated RN hours for all VHA acute care units in CY2020 was 11,088,156 hours. Trends in daily values corresponded with COVID-19 surge periods, similar to that seen in Aim 1a. Across all acute care units, the mean total daily estimated RN hours per acute-care unit was 104.4 hours (SD 64.2). Across all facilities, the mean total daily estimated acute-care RN hours per facility was 961 hours (SD 663).
Aim 1c
We developed a prototype dynamic data display containing staff, patient, and medication pass metrics that can be aggregated at various levels and trended over time and includes data for all VA acute-care units from August 2018 through March 2021. During a demonstration session with 13 MEDVAMC nursing leaders, including care line and nursing managers, they found our peak medpass metrics and prototype dynamic data display helpful for making decisions about staffing and understanding workload on their units.
Dashboard: Design principles
The key deliverable for the project was a dynamic dashboard that provides a summary of staff-, unit-, and facility-level workload, staffing, and imbalance.
The dashboard was designed for two different types of users: (1) A facility-level administrator interested in making facility wide decisions regarding staff resource allocation, and (2) A unit-level manager interested in understanding trends and variability in workload and staffing.
The dashboard was designed to provide trends in workload and staffing across all units within a single facility as well as a detailed drill down within a given unit.
In designing the dashboard, we considered user differences in graph literacy and hence chose to provide visual displays with different levels of complexity.
Key functional features of the dashboard are: 1) An overview page displaying current status of workload and staffing for a single unit; 2) Selectable date range. Short- and long-term trends can be viewed by selecting range of dates over which the data are displayed, at various levels of aggregation; 3) At appropriate levels (unit and facility), over-time aggregates are computed for each metric, with user-defined period of aggregation (1 day, 7 days, 14 days, and 30 days); 4) Selection of facilities and units within each facility; 5) Plots displayed at the facility level and for the unit; 6) Workload computed for all staff as well as by staff role type.
Description of visual elements of the dashboard:
Facility Level: At the facility level: 1) 90-day staffing level for each unit of each unit-type (Critical-care, Step-down, Medical, Surgical, and Mixed Medical-Surgical); 2) A single page with multiple panels displaying trend lines in number of patients and number of staff as well as patients per staff within each unit across the entire year; 3) A single page displaying a stacked bar chart indicating number of occupied beds across all units along with beds occupied by persons under investigation for and confirmed COVID-19 patients.
Unit Level-Daily: Patient Volume, Staffing Levels, and Patients/Staff. A single page that displays boxplots for one of the three metrics (Number of Patients, Number of Staff, Patients per Staff) for a given date range aggregated by a time period (1 day, 7 days, 14 days, and 30 days), by staff type. This display provides a complete real-time picture of adequacy of staffing levels during a given period and allows comparison of trends over time.
Unit-Aggregate: 1) Displays a line chart of trends for one or more metrics (Patients, Staff, Patients per Staff, and Nurse Hours) aggregated over a given period (1 day, 7 days, 14 days, and 30 days), by staff type; 2) Stacked bar chart displaying number of nurses with different levels of patient workload over time.
Unit-level COVID-19: a stacked-bar chart indicating number of occupied beds across all units along with beds occupied by persons under investigation for and confirmed COVID-19 patients.
Aim 2
Our team communicated regularly with operational partners through an email newsletter and virtual meetings. We also shared progress and solicited input at invited presentations. Our team's interviews with stakeholders revealed that many nursing leaders most value real-time and future-forecasted data during emergencies. Operational partners are enthusiastic about this work and our team has plans for further partnership with nursing and hospital medicine leaders.
We developed a newsletter to inform our operational partners of project progress. We received feedback from five issues of this newsletter sent to 46 subscribers, and plan to distribute one to two additional issues. We recruited some newsletter subscribers from the SAIL Governance Group (SGG), which our principal investigator is a member of and attends meetings for monthly. Other newsletter recipients signed up after our team reached out directly to our existing nursing contacts or after word of mouth introduction by their colleagues.
We also received feedback relating to our project through direct conversation with 16 operational leaders about the project and its added value to VA nursing, including direct calls with the VA Chief Nursing Officer, multiple representatives from Nursing Informatics and the Office of Nursing Services, VISN 12 Chief Medical Officer, and VHA Hospital Medicine National Program leaders, and a prototype demonstration with MEDVAMC nursing leaders.
Our team was invited to present for VA Nurses' Week at MEDVAMC in May 2021, and the presentation was given by our co-investigator and project manager. Our team was invited to participate in the ORD COVID Modeling Workgroup. Our team (PI, co-investigator, project manager) also gave an invited presentation about our project at the COVID Modeling Workgroup in June 2021. The presentation included preliminary data on modeling variation in the peak-time medication pass duration using predictors such as number of patients, number of medications, and patient relative risk score.
Our principal investigator made a presentation on a previous facility complexity peer grouping project in December 2019, before the project was initiated, to the SGG; this presentation increased awareness of our work among SGG members and ultimately led to the appearance of our team leaders on COVID in 20 by recommendation of the SGG co-chair. We had the opportunity to connect and met with a program manager from the Office of Nursing Services because of our appearance on COVID in 20, who also referred us to meet with her and her colleagues in late May to discuss the possibility of integrating our work with VA operations nationally. Additional meetings to discuss continued partnership with ONS are planned (next scheduled meeting in late June).

IMPACT:
Monitoring and modeling of temporal and regional variation in direct-care nurse staffing and workload can support (1) deployment of resources to areas of greatest need for Veterans and (2) examination of the impact of varying staffing patterns on outcomes. Tracking changes in staffing levels and workload will be essential to developing appropriate responses to pandemics and other catastrophic stresses to the health care delivery system.

PUBLICATIONS:

None at this time.