SELECT icustays.first_careunit,
caregivers.label AS caregiver_type,
COUNT(*) AS weight_doc_count
FROM chartevents
INNER JOIN caregivers
ON chartevents.cgid = caregivers.cgid
INNER JOIN icustays
ON chartevents.icustay_id = icustays.icustay_id
WHERE chartevents.itemid = 763
AND chartevents.valuenum > 0
GROUP BY icustays.first_careunit, caregivers.label
HAVING weight_doc_count > 0Analysis Report Three - Information Systems and Healthcare Provider Interactions
Executive Summary
Healthcare information systems are meant to support providers, but they can also create frustration when they do not match the realities of clinical workflow. EHRs, clinical decision support tools, and AI can improve access to information and support better care, but they can also increase workload, reduce trust, create workarounds, and make clinicians feel like technology is driving decisions instead of supporting them (McAlearney et al. 2015; Rathert et al. 2019; Bannon 2023).
This analysis uses ICU weight documentation and dietitian-entered nutrition assessment data to show how those larger provider concerns appear in the EHR. The visualizations highlight inconsistent caregiver role labels, duplicated measurement fields, variation across ICU units, and gaps in structured nutrition documentation. These issues may seem small at the charting level, but they reflect larger informatics problems: unclear workflows, multiple places to enter similar information, limited standardization, and data that may be difficult for clinicians and leaders to trust.
For healthcare administrators, the key takeaway is that improving information systems is not only about adding more technology. It is about making the EHR, dashboards, decision support tools, and future AI systems fit the way clinicians actually work. Organizations should standardize caregiver role labels, create one reliable location for daily weight documentation, review ICU documentation workflows, improve structured dietitian fields, and build nutrition screening into routine ICU processes. Before expanding dashboards, alerts, or AI tools, hospitals need to make sure the underlying data is accurate, consistent, and usable for the providers who depend on it.
Introduction
Healthcare technology works best when it helps clinicians make informed decisions without making their work harder. Electronic health records, clinical decision support tools, and artificial intelligence can make healthcare more efficient, connected, and data-driven. However, these tools may also introduce new challenges when data is incomplete, systems are difficult to use, or technology is prioritized over clinical expertise.
Electronic health record implementation is not just a technical project. It is also a major organizational change. Moving from older workflows to an EHR changes how clinicians document care, communicate with one another, and complete daily patient care tasks. McAlearney et al. describe EHR adoption as a change process that can feel like a loss for clinicians, especially when old workflows, professional comfort, and established routines are replaced by new documentation expectations and technology-driven processes (McAlearney et al. 2015). Even a strong EHR system will struggle to improve care if clinicians do not trust it, understand it, or feel supported while using it.
Clinicians who have used EHRs for several years often see both the benefits and the frustrations of these systems. EHRs can make patient information easier to access, help providers review history and trends, and improve communication across the care team. At the same time, they can increase workload, create frustration when systems do not communicate with each other, and require users to sort through large amounts of information to find what is clinically relevant. Rathert et al. found that physicians and nurses recognized advantages such as easier access to patient data, but also reported challenges related to interoperability, workload, training, workarounds, and trust in EHR information (Rathert et al. 2019). For many users, the issue is not whether the EHR has value. The issue is whether it makes clinical care easier or adds another layer of work.
Trust is another major concern. Clinicians may rely on the EHR every day, but that does not mean every piece of information in it is accurate, current, or complete. Documentation may be delayed, copied forward, entered differently across units, or missing important context. When that happens, providers may develop workarounds to get the information they need. Those workarounds may help in the moment, but they can also create new risks for consistency and safety. This matters because the EHR is not just a storage system. It becomes the foundation for quality reporting, clinical decision support, communication, and future analytics.
Artificial intelligence adds another layer to this issue. AI tools can process large amounts of data quickly and may help identify patient risks earlier than humans alone. However, AI should support clinical judgment, not replace it. Bannon describes situations where nurses felt pressure to follow algorithm-driven alerts even when their clinical judgment suggested the patient’s situation was more complex than the AI model recognized (Bannon 2023). Frontline clinicians often notice patient-specific details that an algorithm may miss. When hospital policy makes it difficult to question or override an alert, technology can create moral distress and lead to care that does not fully align with the patient’s clinical picture.
Together, these ideas show that healthcare data is only useful when it is accurate, meaningful, and connected to real clinical practice. EHRs, dashboards, and AI tools all depend on the quality of information entered by frontline caregivers. If documentation is incomplete, inconsistent, or disconnected from workflow, future alerts, reports, and predictive models may be built on weak data.
The Healthcare Context
This tension is especially clear in the ICU. Critical care patients generate an enormous amount of data every day, and clinicians are expected to review, interpret, and act on that information quickly. ICU physicians may monitor approximately 2.5 million data points in a month and respond to an average of 187 EHR alerts per patient per day (Jalilian and Khairat 2022). That volume can easily become overwhelming. When EHR screens are crowded, repetitive, or hidden under layers of menus, finding what matters at the right time becomes harder, not easier.
The problem is not just the amount of data. It is also how the data is displayed and whether clinicians can trust it. Physicians have reported that finding information in the EHR can be challenging because of data-heavy screens, redundant information, and inconsistent values across different areas of the chart (Jalilian and Khairat 2022). In a time-sensitive ICU environment, that matters. If the same patient information appears in multiple places or is not updated consistently, clinicians may spend valuable time determining which value is correct rather than focusing on the patient in front of them. These usability issues can also add to workload, fatigue, burnout, and patient safety risk (Jalilian and Khairat 2022).
Data quality creates another layer of concern. In critical care, missing or inconsistent EHR data can affect clinical decision-making, patient safety, and the performance of machine learning models. A systematized review of critical care EHR data quality found that missing data rates exceeded 80% for some variables, and that missingness was often meaningful rather than random (De Andrade et al. 2026). In other words, what is missing from the chart may reflect clinical workflow, ordering patterns, staffing burden, or documentation habits. Copy-and-paste documentation, delayed charting, inconsistent entries, and workarounds can also weaken trust in the EHR, especially in the ICU, where patients can change quickly (De Andrade et al. 2026).
Daily weight is a strong example of how a basic clinical data point can become an informatics issue. Patient weight is used for medication dosing, fluid assessment, nutrition assessment, skin integrity risk, safe equipment use, and care planning. Yet obtaining and recording an accurate weight remains difficult in day-to-day hospital practice. Barriers include workload, patient acuity, access to equipment, inconsistent measurement conditions, and documentation in multiple areas of the chart (Evans 2012). More recent work also shows that physicians and nurses recognize the importance of accurate intake, output, and body weight documentation, but do not always view the data as reliable (Tao et al. 2024).
In the ICU, daily weight is even more complicated because critically ill patients frequently experience fluid shifts, edema, diuresis, resuscitation fluids, wounds, drains, and changes in lean body mass. A change in weight may not always reflect true nutritional change, but it still remains a key data point for interpreting the patient’s overall condition. From a nutrition perspective, ICU patients are difficult to assess because traditional nutrition risk tools do not consistently identify the same patients, and weight-related data can be affected by fluid status and severity of illness (Coltman et al. 2015).
This also makes the dietitian’s role important. ICU dietitians help identify nutrition risk, estimate energy and protein needs, recommend enteral or parenteral nutrition support, monitor nutrition adequacy, and communicate nutrition plans to the care team. Dietitian-led interventions in critical care have been associated with improved nutrition delivery, earlier initiation of enteral nutrition, and better energy and protein provision (Terblanche et al. 2025). These responsibilities depend on reliable, structured EHR data, including height, weight, feeding weight, estimated energy needs, and estimated protein needs.
This creates the central healthcare context for the analysis. Hospitals are trying to move toward dashboards, AI, and predictive decision support, but those tools depend on the quality of data entered by frontline caregivers. In the ICU, where patients are complex and the pace of care is fast, even basic documentation fields can have major clinical meaning. Before healthcare organizations can fully trust advanced tools, they need to understand whether key data is being documented consistently, where it is being documented, and which caregiver roles are entering it.
Data Visualizations
Visualization One
Daily Weight Documentation by ICU Unit and Caregiver Type
The first visualization shows how often daily weight was documented in each ICU unit and which caregiver types entered the information. Instead of only counting weight entries, this approach uses the CAREGIVERS table to connect each entry to the caregiver role behind it.
The process began by identifying the item ID for daily weight in the MIMIC-III database. Since CHARTEVENTS uses item IDs rather than plain-language labels, the weight-related fields had to be searched first. Item ID 763 was identified as the daily weight field used for this analysis.
CHARTEVENTS was used as the main table because it contains the actual daily weight documentation events. The table was joined to CAREGIVERS using cgid to identify the caregiver type associated with each entry. ICUSTAYS was also joined using icustay_id, which linked each weight documentation event to the patient’s ICU unit.
The WHERE clause filtered the data to include only daily weight records with values greater than zero. This helped remove blank, missing, or unusable entries. The query then used COUNT(*) to count the number of documentation events and grouped the results by ICU unit and caregiver type. The final query asks: for each ICU unit, how many daily weight documentation events were entered, and which caregiver types entered them?
ggplot(data = weightcounts,
aes(x = first_careunit,
y = weight_doc_count,
fill = caregiver_type)) +
geom_col() +
labs(title = "Weight Documentation by ICU Unit and Caregiver Type",
x = "ICU Unit",
y = "Number of weight documentation events",
fill = "Caregiver Type")
The graph indicates that nurses entered the majority of daily weight documentation events. Most caregiver labels appear as RN, but some entries are Rn, and at least one category lacks a clear caregiver label. This is a small but important data quality finding. If the same caregiver role is documented with different capitalization, or if some caregiver IDs lack clear labels, the data becomes harder to summarize and trust. Leaders should consider whether this reflects a simple labeling issue or a broader problem with role mapping, documentation standards, or data governance.
The graph also shows differences across ICU units. MICU had the highest number of daily weight documentation events, followed by SICU, while TSICU had far fewer. This does not automatically mean one unit is performing better or worse. The chart counts documentation events, not unique patients, and it does not adjust for census, patient acuity, or length of stay. However, these differences raise important questions. If certain units have longer ICU stays but fewer daily weight entries, leaders may need to review whether weights are being captured consistently, whether staff recognize the importance of ongoing weight monitoring, or whether this information is being recorded somewhere else in the EHR.
From a workflow perspective, the graph suggests that daily weight documentation may be concentrated among nurses. If this responsibility falls primarily on nursing staff, it could add to documentation burden in an already demanding ICU environment. Leaders should evaluate whether patient care technicians could support weight documentation and whether dietitians, physical therapists, or speech-language pathologists have access to update weight fields when appropriate. If those disciplines document weight-related information somewhere else, the organization may need to review where that data lives and how it is used.
This visual reinforces that daily weight documentation is both a clinical workflow issue and a data quality issue. Weight supports multiple aspects of care, but it only becomes useful for reporting, dashboards, and decision support when it is documented consistently and linked to clear caregiver roles. Without that foundation, future analytics or AI tools may produce incomplete or misleading results (De Andrade et al. 2026).
Visualization Two
Dietitian-Documented Nutrition Assessment Elements by ICU Unit
The second visualization examines nutrition-related documentation entered by dietitian-type caregivers in the ICU. The goal was to identify which nutrition assessment elements were documented by dietitian-related roles and how those patterns varied across ICU units.
The process started with the CAREGIVERS table to understand how MIMIC identifies dietitians. This was not as simple as searching for one label. The CAREGIVERS table included several dietitian-related labels, including RD, DI, DietIn, RD Int, RD/LDN, RD,LDN, and MS,RD. These labels were used to identify charting events entered by nutrition-related professionals.
CHARTEVENTS was used because it contains the actual charting events and includes the caregiver ID attached to each entry. By joining CHARTEVENTS to CAREGIVERS, the query could connect nutrition-related entries back to the type of caregiver who entered them. D_ITEMS was then added because CHARTEVENTS stores charted fields using item IDs, while D_ITEMS provides plain-language descriptions of each field. This helped identify nutrition-related fields, including Estimated Energy Needs/Kg, Estimated Protein Needs/Kg, Feeding Weight, Admission Weight, Height, and Height (cm). ICUSTAYS was also joined, so each entry could be compared across ICU units.
The final query asks: for each ICU unit, how many nutrition-assessment fields were documented by dietitian-type caregivers?
SELECT icustays.first_careunit,
d_items.label AS charted_item,
caregivers.label AS caregiver_type,
COUNT(*) AS chart_count
FROM chartevents
INNER JOIN caregivers
ON chartevents.cgid = caregivers.cgid
INNER JOIN d_items
ON chartevents.itemid = d_items.itemid
INNER JOIN icustays
ON chartevents.icustay_id = icustays.icustay_id
WHERE (caregivers.description = "Dietitian"
OR caregivers.label = "RD"
OR caregivers.label = "DI"
OR caregivers.label = "DietIn"
OR caregivers.label = "RD Int"
OR caregivers.label = "RD,LDN"
OR caregivers.label = "RD/LDN"
OR caregivers.label = "MS,RD")
AND (d_items.label = "Estimated Energy Needs/Kg"
OR d_items.label = "Estimated Protein Needs/Kg"
OR d_items.label = "Feeding Weight"
OR d_items.label = "Admission Weight (Kg)"
OR d_items.label = "Admission Weight (lbs.)"
OR d_items.label = "Height"
OR d_items.label = "Height (cm)")
GROUP BY icustays.first_careunit, d_items.label, caregivers.label
HAVING chart_count > 0ggplot(data = rd_assessment_by_unit,
aes(x = first_careunit,
y = chart_count,
fill = charted_item)) +
geom_col() +
labs(title = "Dietitian-Documented Nutrition Assessment Elements by ICU Unit",
x = "ICU Unit",
y = "Number of charting events",
fill = "Nutrition Assessment Element")
The graph indicates that dietitian-documented nutrition assessment fields were most frequently recorded in the MICU, followed by the SICU. Fewer charting events appeared in the CCU and TSICU, and no CSRU documentation was identified in this output. This does not necessarily mean dietitians were not involved in those units. Documentation may have been entered elsewhere, under different caregiver roles, or in fields not captured by this query. That uncertainty is part of the informatics issue. When data is difficult to locate or scattered across multiple areas of the EHR, it becomes harder for leaders to evaluate nutrition care consistently.
The graph also shows several data standardization issues. As seen in the first visualization, caregiver labels for dietitians are inconsistent, appearing as RD, DI, DietIn, RD Int, RD/LDN, RD,LDN, and MS,RD. While these labels may be familiar to people within the organization, they complicate analysis. When the same role is entered in multiple ways, filtering, summarizing, and trusting the data becomes more difficult. Organizations relying on this information for reporting or dashboards need clearer role mapping and stronger data governance.
The nutrition assessment fields also show duplication in measurement units. Weight-related fields appear as Admission Weight (Kg) and Admission Weight (lbs.), while height appears as Height and Height (cm). These differences matter because structured data is easier to analyze when it is standardized. Without reliable units, conversion rules, and clear documentation standards, downstream tools may misread, duplicate, or exclude important information.
One important finding is that estimated energy and protein needs appear relatively low compared with the other fields. In ICU nutrition care, energy and protein needs are central to assessment and planning. Critically ill patients are at high risk for malnutrition, and nutrition risk assessment in the ICU can be difficult because traditional screening tools do not always identify the same patients (Coltman et al. 2015). Dietitians play an important role in setting nutrition targets, developing feeding plans, monitoring adequacy, and supporting nutrition delivery in critical care (Terblanche et al. 2025).
The lower number of calorie and protein need entries should be viewed as a documentation and workflow signal, not evidence that dietitians are not assessing patients. Leaders would need to review whether those estimates are documented in other locations, whether the consult process is consistent, whether ICU nutrition screening is built into the workflow, and whether dietitians have access to the right structured fields.
From a staffing and workflow perspective, this visualization prompts leaders to consider whether dietitian coverage aligns with ICU patient needs. If critically ill patients require nutrition assessment, feeding plans, and monitoring, but structured documentation is limited in certain units, leaders may need to review staffing levels, consult volume, unit expectations, and documentation standards. This reinforces that dietitian documentation is more than nutrition charting. It is also a workflow, data quality, and governance issue. Fields such as feeding weight, height, estimated energy needs, and estimated protein needs help turn clinical assessment into usable data. This shows why standardized documentation is needed before nutrition-related dashboards, alerts, or AI tools can be fully trusted.
Recommendations for Industry
The research and data findings show that healthcare organizations cannot treat EHR documentation as a routine back-end task. Documentation quality affects clinical decisions, workflow efficiency, reporting, dashboards, and future AI tools. For administrators, the goal should be to improve how EHR data is captured, standardized, displayed, and used in real ICU workflows.
Standardize caregiver role labels
Both visualizations revealed inconsistent caregiver labeling. In the daily weight analysis, nurse labels appeared as both RN and Rn. In the dietitian analysis, dietitian-type caregivers were identified under several labels, including RD, DI, DietIn, RD Int, RD/LDN, RD,LDN, and MS,RD. While these differences may appear minor, they complicate the evaluation of documentation patterns, staffing responsibilities, and quality performance.
Healthcare administrators should work with clinical informatics, human resources, and EHR analysts to map similar caregiver roles into consistent categories. Dietitian-related labels should roll up into one standardized dietitian category, while nursing labels should be consistently mapped to RN or another agreed-upon role. If the organization cannot clearly identify who entered the data, it becomes harder to evaluate workflows or trust reports.
Standardize daily weight documentation and review ICU workflow
The first visualization identified variation in daily weight documentation across ICU units and caregiver types. Most daily weights were entered by nurses, but some units recorded fewer documentation events. This raises questions about consistency, documentation location, and workflow expectations across units.
Healthcare organizations should establish one standard EHR location for daily weight documentation and clearly define when and how daily weights should be updated. Weight should not be scattered across multiple screens, flowsheets, or unit-specific charting locations unless those fields are clearly connected behind the scenes. A standardized daily weight workflow would improve trust in the data and help clinicians, dashboards, and decision support tools work from the same source of information.
Administrators should also review daily weight documentation by ICU unit rather than relying only on aggregate totals. Variation may be influenced by census, acuity, length of stay, patient type, or workflow, but it still warrants review. This review should involve nurses, patient care technicians, dietitians, and unit leaders. If nurses are responsible for most daily weight documentation, leaders should assess whether this responsibility is appropriate, whether it increases nursing workload, and whether certain tasks could be shared with patient care technicians.
Improve structured dietitian documentation and nutrition screening workflows
The second visualization showed that dietitians documented estimated energy and protein needs less frequently than height- and weight-related fields. This does not mean dietitians were not assessing ICU patients. However, when key nutrition data is recorded in narrative notes, consult notes, flowsheets, or other fields not included in reporting, the organization loses value from that documentation.
Healthcare administrators should work with dietitians and EHR analysts to improve structured nutrition documentation fields. Estimated energy needs, estimated protein needs, feeding weight, nutrition route, nutrition goals, and nutrition adequacy should be captured in consistent fields when possible. These fields should be easy for dietitians to access and should connect to reporting tools so leaders can evaluate whether ICU patients are receiving timely nutrition assessment.
Leaders should also review how dietitian involvement is initiated in the ICU. If a dietitian assessment depends on provider consults, inconsistent consult practices may result in inconsistent nutrition documentation. This is likely a workflow issue rather than a reflection of dietitian performance. ICU nutrition screening should be built into admission and daily review processes, especially for patients with mechanical ventilation, enteral or parenteral nutrition, prolonged NPO status, significant weight change, wounds, or extended ICU stays.
Design EHR fields around clinical workflow
The research shows that EHR usability is a major issue in the ICU. ICU clinicians manage large volumes of patient data, and crowded or repetitive screens can make it harder to quickly find the information they need. Administrators should assess whether daily weight, feeding weight, estimated energy and protein needs, and nutrition plans are easy to find within the EHR.
The goal is to reduce duplicate documentation, remove unnecessary steps, and present nutrition-related data in a consistent location. When fields are difficult to find, duplicated, or inconsistently named, clinicians may create workarounds that weaken data reliability over time. A better-designed workflow should make accurate documentation easier, reduce confusion about where data belongs, and support the clinicians who need to act on that information.
Audit data before expanding dashboards or AI tools
Before expanding nutrition dashboards, malnutrition alerts, or AI-supported decision tools, organizations should complete foundational data audits. These audits should identify missing values, duplicate fields, inconsistent units, unclear caregiver labels, and documentation differences across units. The current analysis identified each of these concerns.
This should happen before advanced analytics are treated as reliable. If the underlying EHR data is incomplete or inconsistent, dashboards and AI tools may give leaders a false sense of accuracy. For ICU nutrition, ongoing data quality reviews should include daily weight, feeding weight, height, estimated energy and protein needs, route of nutrition, and dietitian involvement. These reviews should involve both data analysts and frontline clinicians to determine whether missing fields reflect technical issues, workflow challenges, or actual gaps in care.
These recommendations are not about adding more documentation. They focus on making documentation more reliable, standardized, and useful so the EHR better reflects real patient care. If healthcare organizations want to rely on dashboards, clinical decision support, and AI tools, the data behind those tools must be accurate, consistent, and connected to the workflows of the people entering it.
References
Bannon, Lisa. 2023. “When AI Overrules the Nurses Caring for You.” The Wall Street Journal 15.
Coltman, Anne, Sarah Peterson, Kelly Roehl, Hannah Roosevelt, and Diane Sowa. 2015. “Use of 3 Tools to Assess Nutrition Risk in the Intensive Care Unit.” Journal of Parenteral and Enteral Nutrition 39 (1): 28–33.
De Andrade, Joao Brainer Clares, Marconny Alexandre Oliveira de Medeiros Cavalcante, Thiago Luı́s Marques Lopes, João Marcos Secundino Treigher, Mateus Dutra Balsells, Júlia Lima Vasconcelos, Lis Cavalcante Monteiro, and Déborah Danna da Silveira Mota. 2026. “Discovery of Data Quality Issues in Electronic Health Records: Profound Consequences for Critical Care Medicine Applications–a Systematized Review.” Critical Care 30 (1): 19.
Evans, Alison. 2012. “Positive Patient Outcomes in Acute Care: Does Obtaining and Recording Accurate Weight Make a Difference?” The Australian Journal of Advanced Nursing 29 (3): 62–67.
Jalilian, Laleh, and Saif Khairat. 2022. “The Next-Generation Electronic Health Record in the ICU: A Focus on User-Technology Interface to Optimize Patient Safety and Quality.” Perspectives in Health Information Management 19 (1).
McAlearney, Ann Scheck, Jennifer L Hefner, Cynthia J Sieck, and Timothy R Huerta. 2015. “The Journey Through Grief: Insights from a Qualitative Study of Electronic Health Record Implementation.” Health Services Research 50 (2): 462–88.
Rathert, Cheryl, Tracy H Porter, Jessica N Mittler, and Michelle Fleig-Palmer. 2019. “Seven Years After Meaningful Use: Physicians’ and Nurses’ Experiences with Electronic Health Records.” Health Care Management Review 44 (1): 30–40.
Tao, Jianling, Sara May, Mingyi Li, Marianne Monahan, Donna Phanumas, and Charles Seelig. 2024. “A Survey for Charting Intake, Output, and Body Weight in the Electronic Medical Record.” The Journal for Healthcare Quality (JHQ) 46 (5): 293–99.
Terblanche, Ella, Judith Merriweather, Mina Bharal, Rhiannon Lewis, and Liesl Wandrag. 2025. “Dietetic-Led Interventions in Critically Ill Adults That Influence Outcome: A Scoping Review.” Journal of Parenteral and Enteral Nutrition 49 (2): 165–79.