Quantifying and Visualizing Nursing Flowsheet Documentation Burden in Acute and Critical Care

Abstract

Documentation burden is a well-documented problem within healthcare, and improvement requires understanding of the scope and depth of the problem across domains. In this study we quantified documentation burden within EHR flowsheets, which are primarily used by nurses to document assessments and interventions. We found mean rates of 633-689 manual flowsheet data entries per 12-hour shift in the ICU and 631-875 manual flowsheet data entries per 12-hour shift in acute care, excluding device data. Automated streaming of device data only accounted for 5-20% of flowsheet data entries across our sample. Reported rates averaged to a nurse documenting 1 data point every 0.82-1.14 minutes, despite only a minimum data-set of required documentation. Increased automated device integration and novel approaches to decrease data capture burden (e.g., voice recognition), may increase nurses’ available time for interpretation, annotation, and synthesis of patient data while also further advancing the richness of information within patient records.

Introduction

Clinical data capture and documentation should be: clinically pertinent, of high quality, efficient and usable, support multiple downstream uses as a byproduct of recording care delivery, enable shared decision-making and collaboration, enable collection and interpretation of information from multiple sources, and be automated whenever appropriate.¹ Decreasing documentation burden across healthcare settings and professionals is a priority of several professional organizations, government agencies, and applied informatics efforts.^1-4 AMIA’s report on the EHR 2020 Task Force’s first recommendation was to decrease documentation burden.⁴ Nurses are one of the largest users of health information technology and have some of the highest levels of documentation requirements within acute care settings. EHR design should balance capturing data to promote safety, quality, and high reliability of care, as well as patient engagement and satisfaction, while minimizing “death by data entry” for all health professionals.³ Focusing on minimizing “death by data entry” - while expanding the use of EHRs as a cognitive support tool and platform for clinicians to consume data and interpret information - is particularly important for those professions that suffer from a high documentation burden, such as nurses and physicians.

Several studies have described the problem of documentation burden, including time-motion studies, reporting that nurses spend between 19% to 35% of nursing practice time documenting care.^5-7 However, there are inconsistent findings related to the impact of EHRs on time spent charting.^8,9 Time motion studies provide deep understanding of clinical system use in practice, work patterns, workload, and time allocation for discrete activities (such as documenting clinical care) to help identify workflow inefficiencies and define optimal workflow solutions.^10,11 However, time motion studies require significant resources and are limited by the time period observed and the validity and reliability of the data collection methods.^5,12-17 Furthermore, in time motion studies it is challenging to differentiate time spent for data consumption versus data entry, with the former aligned with use of the EHR as a cognitive support tool and the later risking use of the clinicians time as a “data entry clerk”. In 2015, O’Brien et al noted that “physiologic monitors, ventilators, low acuity vital signs machines, anesthesia machines, and other point of care devices are rarely fully integrated with the EHR, requiring nurses to manually enter electronic device data into EHR”.³ In response, our team sought to utilize log-file analyses to understand, quantify, and visualize the problem of documentation burden for a specific use case: nurses’ flowsheet data entries in acute and critical care units.

Few studies have investigated documentation burden by quantifying the number and frequency of data points entered into an EHR using analyses of data entry log-files. One recent study queried EHR log-files to determine the total note documentation entries by trauma surgeons in 2014 compared to hospital charges and work relative value units (WRVU).¹⁸ This study concluded that trauma surgeons experience a higher documentation burden for defining hospital charges and WRVU compared to orthopedic surgeons and neurosurgeons. Hripcsak et al., performed a log-file analysis of clinicians’ time entering and viewing notes and reported that nurses spend 21.4 - 38.2 minutes per day authoring narrative notes, on average.¹⁹ However, as noted by Hripcsak et al., the analyses did not include flowsheet documentation, which includes vital signs, physiological assessments, intake and output, treatment parameters (isolation, wound care, etc.), and data for quality initiatives such as falls risk assessment.¹⁹

Aligned with sentiment expressed in AMIA’s EHR 2020 Task Force report, flowsheet data elements that nurses are expected enter should be supported by evidence and should support nurses’ clinical decision making and team-based decision-making, rather than regulatory requirements or non-clinical operations.⁴ This study explores the documentation burden in flowsheets, which span the breadth of clinical assessments and interventions performed by a nurse. Further, we performed a sub-analysis of device integrated flowsheet data to understand current state of device integration and opportunities for greater automated data streaming that will allow nurses to practice at the highest level of their degree by engaging in data interpretation, annotation, and synthesis activities, rather than data entry activities.

Methods

Our team queried all flowsheet data entries on 4 acute care general medicine units (GMU) and 2 medical intensive care units (MICU) for 12 months (January 2017 – December 2017) at a large academic medical center that used a vendor EHR in the Northeastern United States. For each flowsheet data entry, we captured: patient identifier, patient encounter identifier, flowsheet template, flowsheet data element name, free text comment associated with flowsheet data entry, date and time of the data entry, user identifier, and the clinical unit. The flowsheet documentation at the institution in this study included a minimum data set of required data elements, such as vital signs and pain assessment, with the majority of the flowsheet data elements determined by the nurse as appropriate for the individual patient case.

Given our aim of understanding structured documentation burden for bedside nurses on our study units, we excluded any data entered by users that were not registered nurses or licensed practical nurses, such as patient care assistants. We chose not to calculate the data entered per patient since there could be instances when more than one nurse was caring for a patient, such as a critically unstable ICU patient. We also chose not to analyze our data set for total and mean data entries per nurse since nurses may work a varied number of shifts per week and the length of their shifts may also vary overtime. Therefore, we chose to quantify the mean number of flowsheet data entries per nurse who worked a 12-hour day shift (07:00 – 18:59) and who worked a 12-hour night shift (19:00 – 06:59), since 12-hour shifts are typical in many acute and critical care settings. Using one standard time range allowed us to avoid including nurses that worked shorter shifts in our denominator of nurses working on a given day which would artificially decrease the mean number of data entries per nurse on a unit. To exclude nurses that worked less than 12 hours we subtracted their first data entry from their last data entry on a given day; if that number was less than 9 hours we determined it was unlikely the nurse worked a 12-hour shift. For example, it is possible during a busy shift a nurse may continue to enter documentation into the next hour (up to 9 hours for an 8-hour shift), but unlikely that a nurse would not enter any data for more than 1.5 hours on either end of his/her shift (less than 9 hours for a 12-hour shift). We quantified the number of flowsheet data points documented during our defined 12-hour shifts on each unit and divided it by the number of nurse users that documented during that same time period to determine the mean number of data entries per nurse. For each time period on each clinical unit, the following calculations were performed:

• a)Mean number and standard deviation of data points documented
• b)Mean number and standard deviation of users that documented
• c)Mean number of data points per user

We also performed a sub-analysis of all the flowsheet data entries that were auto-populated based on device integration, such as vital sign and cardiac monitors. There were 61 device integrated data elements out of 4198 total data elements available in the EHR flowsheets. Importantly, these auto-populated fields still required the nurse to manually review and validate the accuracy of the data once it appears in the flowsheet view before it is saved as ‘validated’ to the patient’s record, hence, decreasing risk for transcription error but not limiting the documentation burden entirely. The data point we used in our analysis for the device data was the date-time stamp from when the nurse validated it. Finally, to understand the temporal nature of flowsheet documentation over a 12-hour shift we randomly selected 3 use cases from each shift (day and night shift) and type of clinical unit (acute and critical care) in our data set, for a total of 12 use cases, and plotted them on a timeline with the timestamp for each data entry on the x-axis. Institutional Review Board (IRB) approval was obtained for this study.

Results

We found that, on average, nurses perform 787 - 852 flowsheet data entries per 12 hour shift in an ICU and 667 - 930 flowsheet data entries per 12 hour shift on an acute care floor. Overall, rates appeared to be similar between the two ICUs and between the four acute care units, with the exception of the acute care units during the night shift. Rates during the night shifts were higher for each unit than the corresponding rates during the day. In the acute care units, nurses care for 4 patients at night and 3 patients during the day, on average.

Device integrated data accounted for 5-20% of flowsheet data entries across our sample. As expected, there was a greater portion of device integrated data captured in the ICU than in the acute care units. After excluding the device integrated data points, nurses in both the ICU and acute care units were found to document in the range of 631 -662 data points per 12-hour day shift. During a 12-hour night shift, the range of manually entered data points was 680-689 in the ICU and 728-875 in the acute care units.

These reported rates average to 1 data point every 1.04-1.14 minutes in the ICU and document 1 data point every 0.82 – 1.14 minutes in acute care, after excluding device data. The visualizations in Figures 1 and 2 support that short intervals, measured in minutes, exist between a nurses’ flowsheet data entry activities. See Figures 1 and 2 for counts per time stamp entry, per unit type and shift. The 12 use cases, 3 from each shift per unit type, are plotted on a timeline with the timestamp (in hours and minutes) for each data entry on the x-axis and the count of data points entered at the time on the y-axis. Overall, these visualizations indicate that nurses may enter up to 40 flowsheet data points with relative frequency throughout their shift and may enter a larger batch of data at only one or two time points during their shift.

Figure 1.

Example Cases of Number of Data Entries per Time Stamp for ICU and Acute Care 12 Hour Day Shifts

Figure 2.

Example Cases of Number of Data Entries per Time Stamp for ICU and Acute Care 12 Hour Night Shifts

Discussion

Our findings that nurses document an average range of 631-875 manually entered flowsheet data points per shift (excluding device data), are complementary to other data reported in the literature that concluded that nurses spend between 19% to 35% of nursing practice time documenting care.^5-7 In our study we chose to quantify the amount of data rather than time because we believe that quantifying documentation burden using complementary metrics (time and data quantity) will provide a more complete picture of the problem and assist in targeting specific functionalities and content domains within the EHR for improvement. Importantly, nurses also perform other types of EHR documentation beyond our focus in this study, including: medication administration, plan of care and patient education documentation, and several types of narrative notes (e.g., progress notes, significant event notes, transfer notes, discharge notes). For example, Hripcsak et al in an analysis of EHR log files found that nurses spend 21.4 - 38.2 minutes per day authoring narrative notes on average.¹⁹ This note writing time is in addition to the time it takes a nurse to: document 631-875 flowsheet data points per 12-Hour shift and record medications administered, document patient education performed, develop and document a patient’s plan of care, review the patient’s historical and current data, read team notes, read and send electronic communications, and prepare the patient’s discharge. Notably, this list is not exhaustive of all nursing activities within the EHR. Therefore, our targeted findings should be interpreted as only one portion of nurses’ total documentation burden.

As stated previously, the flowsheet documentation in this study included only a minimum data set of required data elements. Therefore, our findings should not be interpreted as evidence that organizations require high levels of flowsheet documentation. Rather, we believe our findings identified the extensive amount of data and information gathered by nurses during the process of patient care delivery, and that informatics tools should support more efficient data capture so that nurse can spend more time interpreting, annotating, synthesizing and communicating these patient data. Figures 1 and 2 provide a partial snapshot of a nurses’ day, in the context of flowsheet usage. The instances of observed high number of data entries per time stamp could be interpreted in several ways, for example: the nurse is busy and has collected more data (e.g., new patient admission, patient transfer for procedure) or the nurse is less busy and is afforded more time to document. In addition, we note that our data showed higher mean rates of documentation during the night shift. While no concrete conclusions can be drawn from these limited visualizations and data, they suggest that further work is needed to understand when and how nurses choose to document, and how those choices impact signals and/or biases within the data that are essential to understand for analytical purposes.²⁰ Our team has identified these types of signals in prior work, where increased patterns of flowsheet documentation served as a proxy of a nurse’s concern about the patient’s risk of decompensation and were associated with rates of mortality and cardiac arrest.^3,11

O’Brien and colleagues noted in 2015 that “nursing content is often conceptualized as though it occurs on paper with limited considerations about how it supports nursing-based decision making or patient engagement”.³ We recommend deeper analyses into the use, and potential use, of these data points documented by nurses to help redesign data collection that, as articulated by Payne et al. in 2015, should only include those data that are necessary to diagnose and treat a patient’s condition (we note: inclusive of nursing diagnoses and interventions) and do not add to the documentation burden.⁴ Innovative tools should automate data capture through devices, voice, or similar modalities and support nurses’ information generation and interpretation for nursing diagnoses, nursing interventions and nursing plans of care; relevant interdisciplinary data for medical diagnoses, treatment, shared decision-making, communication and planning; nursing sensitive outcomes; and patient centered outcomes.

Our data indicate there is room to expand device integration opportunities to ‘free-up’ time for nurses to perform data interpretation, annotation, and synthesis that provides context for structured data to ‘tell the story’, rather than spending time performing data entry. Payne et al., in 2015, noted that important aspects of patient stories can only be effectively captured by narratives⁴ ; our team’s prior work indicated that nurses document short narrative comments in flowsheets to contextualize the structured data as a means to communicate the essence of the patient’s status at that point in time.^20,21 Payne et al., also recommended that “policymakers should encourage fully standardized interfaces between IT systems, as opposed to requiring users to manually transfer clinical data between separate medical devices or other external sources”.⁴ Based on our data, nurses are currently documenting in a hybrid model where a minority of the flowsheet data is automated and most is manually entered. We recommend that future investigations evaluate cognitive load during use of a hybrid model for flowsheet documentation, compared to a model for flowsheet documentation that maximizes automated data streaming and voice recognition so that the nurse is able to focus on data interpretation and synthesis, rather than data entry. As healthcare moves to value based care reimbursement, the industry’s standard for data collection methods should similarly push innovative and novel boundaries that provide maximum value to all clinicians, and ultimately patients, at the point of care.⁴ Our data supports inpatient nursing flowsheet documentation as an important use case for these efforts.

Currently, EHR Flowsheets have several different functionalities or configurations (in addition to device integration) to increase documentation efficiency. While these functionalities arguably fall short of data capture solutions available in other industries (or within our own smartphones), we believe they are worthy of mention for future studies that may attempt to replicate our results and for reference when considering innovative EHR solutions to optimize burden. The first is cascading, which is a logic function that displays new flowsheet data elements to the user if a certain value is entered. This functionality assists in navigation and may decrease information overload, yet does not decrease manual data entry and poses challenges to determine accurate denominators or missing data when performing analytics. The second functionality in many EHR flowsheets is the ability to assert when data is ‘within defined limits’ (i.e., normal per an institution’s definition). This functionality will allow a nurse to assert that data, or an entire group of data, is ‘within defined limits’ and it is possible an EHR may store a value for each data element in batch when supporting this type of workflow. This functionality was not available to users at our study site since it was decided it would limit the ability to perform future outcomes analyses, however, any future log-file analyses should consider if it is present, how it behaves, and how to account for it. Finally, many EHRs allow users to copy data from one flowsheet column to another. This functionality typically copies all data from a column into a new column, requiring a nurse to manually change any information that should not be copied. The EHR we studied did allow users to copy data from one column to another, however, it did not permit copying for any assessment data elements and was limited to data that typically remains static, such as a patient’s social and history data. It is possible that this limited functionality was used within our data set; yet, figures 1 and 2 infer that this functionality was likely not used at a frequent rate given that none of our randomly selected users had counts that were consistently the same as the prior time stamp. Additional temporal analyses may be useful to understand if and how nurses engage in workflows to ‘batch document’ at intervals during clinical shifts.

Additionally, we note the methodological importance of recognizing that the length and frequency of nurses’ shifts vary and the number of patients that a nurse cares for also varies across different settings. In cleaning our data, we identified several 4 hours shifts in addition to the more typical 8 and 12 hour shifts. These 4 hour ‘shifts’ may occur, for example, to provide coverage when a nurse calls out sick or when a ‘Charge Nurse’ on the unit (who typically does not have a patient assignment) provides direct care for a patient for a short period of time due to an increased census on the unit. Excluding nurses that worked less than 12 hour shifts from our sample was an important methodological step to avoid including shorter shifts that naturally had fewer data points entered during them and lowered the average. While this is mathematically obvious, we point it out so that investigators for future studies are aware of the variable lengths of nursing shifts when interpreting EHR log-files. We recommend that future studies of log-file analyses account for the time a nurse actually spent working and directly caring for a patient. Important next steps include replicating our analyses for the 8-hour shifts in our data set and expanding these analyses to include other health professionals. Additionally, further comparisons between units and factors contributing to burden are important areas for future work.

Limitations

This study was limited by use of data from a single academic medical center and from EHR log-files without direct observation. While a time-motion or workflow observational study would provide deeper insights into nursing documentation activities during the time observed, our methods provide the ability to understand overall documentation burden across units and shifts for nurses. This study calculated mean rates of data entered per unit of time within a 12-hour clinical shift rather than reporting temporal trends of data entry (with the exception of the case visualizations). All the flowsheet documentation for a patient during a time period may not be from only 1 nurse due to nurses ‘covering’ for each other while on breaks (e.g., lunch). However, in our experience documentation during coverage for breaks is minimal and is performed by other nurses on the unit who are already factored into the denominator, not an additional set of nurses. As we noted above, users could have copied limited types of data, however, this did not include nursing assessments and our visualization suggests this likely was not a frequent pattern. Finally, it is important to note that our study only analyzed one type of EHR documentation performed by nurses – flowsheet data – and not other types of structured data entry or note writing. This study also only focused on entry of data into flowsheets and was not focused on viewing/consuming data within the EHR.

Conclusion

AMIA’s ‘EHR 2020’ Task Force included as its first recommendation to decrease documentation burden. Documentation burden is a well-recognized issue within healthcare. A greater understanding of the scope and depth of the problem across clinical domains is needed in order to identify areas that are high priority for optimization and are amenable to technological solutions. In this study we quantified nursing documentation burden in the context of EHR flowsheets which are used primarily to document patient assessments and interventions. Our data showed mean rates of 633-689 manual flowsheet data entries per 12-hour shift in the ICU and 631-875 manual flowsheet data entries per 12-hour shift in acute care, excluding device data.

There remains substantial room for decreasing documentation burden, increasing automated data streaming from devices, and supporting data capture through novel approaches such as voice recognition. Interestingly, we found high rates of documentation in the setting of a minimum data set of documentation requirements, suggesting that documentation patterns within nursing flowsheet data may be based on expert nursing practice rather than institutional policy. Our team is currently investigating these flowsheet data patterns using machine learning to identify signals of nurses concern for patients at risk of deterioration in the hospital. Innovative data capture tools that empower nurses as expert data interpreters rather than data entry clerks may further advance the richness of information within patient records. Based on the data presented in this study and supporting literature, we believe that applied informatics research and innovations should aggressively focus on decreasing data entry burden for nurses - and all care team members – and increase automated data streaming through device integration whenever possible and evaluate the impact on care quality, data quality, and data interpretation and clinical decision making at the point of care.

Table 1.

Mean Flowsheet data points per 12 hour DAYTIME shift

Unit Type		Total data points for unit (SD)	Nurses on unit (SD)	Data points/nurse	Device data points/nurse	Manually entered data points/nurse N (%)	Manually entered data points/nurse/hour	1 data point manually documented every:
ICU	Unit A	5531.45 (1210.82)	6.82 (1.20)	810.50	160.31	650.19 (80%)	54.18	1.11 minutes
	Unit B	5522.36 (1168.48)	7.01 (1.16)	787.37	153.93	633.44 (80%)	52.79	1.14 minutes
Acute Care	Unit C	1993.36 (747.02)	2.83 (1.07)	705.37	43.29	662.09 (94%)	55.17	1.09 minutes
	Unit D	2071.64 (838.38)	2.94 (1.17)	704.99	49.13	655.86 (93%)	54.65	1.10 minutes
	Unit E	2524.28 (782.87)	3.70 (1.06)	683.66	39.68	643.98 (94%)	53.67	1.12 minutes
	Unit F	2440.48 (772.50)	3.66 (1.11)	667.25	35.64	631.60 (95%)	52.63	1.14 minutes

^*nurse to patient ratio in ICU is 1:1 to 1:2, depending on patient acuity;

^±nurse to patient ratio in acute care is 1:3 for day shift

Table 2.

Mean Flowsheet data points per 12 hour NIGHTTIME shift

Unit Type		Total data points for unit (SD)	Nurses on unit (SD)	Data points/nurse	Device data points/nurse	Manually entered data points/nurse N(%)	Manually entered data points/nurse/hour	1 data point manually documented every:
ICU* (nurse to patient ratio 1:1 to 1:2, depending on patient acuity)	Unit A	5675.41 (1221.78)	6.66 (1.23)	852.25	163.11	689.13 (81%)	57.43	1.04 minutes
	Unit B	5589.35 (1150.40)	6.70 (1.11)	833.82	153.05	680.77 (82%)	56.73	1.06 minutes
Acute Care±	Unit C	2604.45 (960.50)	2.80 (0.95)	930.53	54.96	875.57 (94%)	72.96	0.82 minutes
	Unit D	2695.73 (930.06)	2.99 (0.99)	901.07	54.23	846.84 (94%)	70.57	0.85 minutes
	Unit E	2467.50 (741.00)	3.20 (0.87)	770.83	42.25	728.58 (95%)	60.72	0.99 minutes
	Unit F	2563.38 (718.18)	3.24 (0.91)	790.07	39.82	750.25 (95%)	62.52	0.96 minutes

^*nurse to patient ratio in ICU is 1:1 to 1:2, depending on patient acuity;

^±nurse to patient ratio in acute care is 1:4 for night shift