A model for increasing patient safety in the intensive care unit: increasing the implementation rates of proven safety measures
- W S Krimsky1,
- I B Mroz2,
- J K McIlwaine2,
- S D Surgenor2,
- D Christian2,
- H L Corwin2,
- D Houston2,
- C Robison2,
- N Malayaman1
- 1Critical Care Medicine, Franklin Square Hospital, Baltimore, Maryland, USA
- 2Critical Care Medicine, Department of Anesthesiology, Dartmouth-Hitchcock Medical Center, Lebanon, New Hampshire, USA
- Dr W S Krimsky, Franklin Square Hospital, 9103 Franklin Square Drive, Suite 300, Baltimore, MD 21237, USA;
- Accepted 28 March 2008
Background: Few published data exist with respect to current implementation of interventions that increase patient safety in intensive care units (ICUs) Furthermore, even less published data exist that address implementation of outcome-related methodologies of patient safety interventions in ICUs.
Objective: The purpose of this study was threefold: (1) to increase implementation rates of known, evidence-based interventions in the Dartmouth Hitchcock Medical Center (DHMC) ICU that have been demonstrated to reduce morbidity and mortality in critically ill patients; (2) to develop a durable and reproducible intervention model that can be applied not only to various aspects of ICU medicine but to any healthcare microsystem that is process oriented; and (3) to design an “ICU-specific” value compass.
Design: Using a before/after study design, the interventions involved: (1) establishing a systematic approach to integrate the delivery of proven ICU safety measures; (2) using the design of the various tools to develop a method for team communication and team building; (3) incorporating prompts into a ICU progress note for the healthcare team to address three evidence-based measures on a daily basis; and (4) using a data wall to demonstrate progress and to provide “real-time” feedback for error correction.
Setting and participants: In the before and after study, two groups of 40 consecutive patients admitted to DHMC’s Intensive Care Unit were evaluated. The first group of patients was admitted between April and May of 2003. The second group of 40 patients was admitted between May and June of 2004. To ensure process stability, control data were also collected on patients at an interval time point between these two groups.
Main outcome measures: Three evidence-based interventions were identified that reduce the likelihood of adverse events resulting simply from an ICU stay: (1) prophylaxis against venous thrombo-embolic disease (venous thromboembolism or deep vein thrombosis); (2) prophylaxis against ventilator-associated pneumonia (VAP); and (3) prophylaxis against stress-ulcers (SU). Two data points were obtained per patient per day corresponding to the work shift schedule in the ICU. The unit of measure was patient-shift observation. A limited data set was collected before implementing the change package to ensure system stability.
Results: Both traditional statistical analysis and statistical process control (SPC) were used to evaluate the results. For each metric, it was possible to demonstrate an increase in the measure of the mean, reduced point-to-point variation as well as a substantial narrowing of the control limits indicating improved process control.
Limitations: By virtue of the involvement of the researcher in the data collection for the control group, the potential existed for methodological bias by acting on the information collected. There was also the lack of a cohesive data structure from which to collect information (ie, the hospital computer speaks one language, the ventilator a second and the monitoring systems a third).
Conclusions: A model for changing the ICU microsystem at DHMC was created that enabled successful implementation of evidence-based measures by maximising the natural flow of work and fostering a team-based culture to improve patient safety. Unique to this method and unlike currently available methods that define only the delivery of the appropriate intervention as success, system success was defined in terms of both true positives, namely delivering care when it is indicated, as well as true negatives, not delivering care when there is none indicated, to offer a more comprehensive system review. Additionally, the method of data collection allowed simplified defect analysis, thereby eliminating a resource-consuming audit of data after the fact. This approach, therefore, provides a basis for adapting and redesigning the PDSA cycle so as to specifically apply this type of “disciplinary” work.
Since the late 1990s, a series of reports released by the Institute of Medicine (IOM) have underscored how the healthcare system in the United States too often fails to provide patients safe and effective care.1–3 Despite the availability of evidence-based interventions that increase patient safety in intensive care units (ICUs), evidence suggests that the data and tools necessary to implement safe practices and interventions are often not being used by frontline healthcare workers.4–6 A number of projects have been initiated recently designed to change the ICU culture and redesign the system of care on a macro level so that interventions are more quickly and uniformly adapted by frontline workers. Several ICU safety and culture improvement projects are currently creating collaborative learning networks among hospitals on a statewide level so that evidence-based care practices are translated to routine frontline care.7–13
Among the effective metrics identified in the literature are three process measures that are particularly efficacious in reducing mortality and morbidity as well as cost.14–16 The measures, now part of The Joint Commission ICU core measures set, are: (1) prophylaxis against venous thromboembolic disease (venous thromboembolism (VTE) or deep vein thrombosis (DVT)); (2) prophylaxis against ventilator-associated pneumonia (VAP); and (3) prophylaxis against stress-ulcers (SU).17 While the literature provides ample evidence of the efficacy of these measures, little published data exist with respect to current implementation rates of these interventions on the microsystem level.17–23 Similarly, no published data exist addressing the methodology of implementation.
This knowledge led to the design of a model to integrate patient safety measures more effectively into the ICU microsystem while simultaneously allowing examination of the implementation methodology. The intervention established a general systematic approach to integrating the delivery of evidence-based medicine into daily rounds in the DHMC ICU in order to improve patient safety. Attending physicians, house officers, nurses, respiratory therapists and pharmacists participated in the daily rounds. This approach enabled the team to design a data-collection strategy to mirror the process of care delivery, thereby providing real-time insights into system defects; enabling real-time correction of errors with the stated goal of reaching “zero-defects” in the delivery of care.
The data-collection tool developed during this study, while adapted for the specific context of an ICU, can easily be applied to any area of healthcare that has a disciplinary focus and is process-oriented, such as perioperative medicine.
This project involved four phases over a 2-year period, and was based on synthesis data drawn from both the Agency for Healthcare Research and Quality (AHRQ), and subspecialty specific guidelines, recommendations and systematic design principles.424–27 The work included the development of an adapted version of the PDSA cycle during the action and testing period. The first phase was to identify defects in the delivery of evidence-based practice grounded in Batalden’s model of microsystem care implementation (fig 1). This model illustrates the three generally accepted components regarding implementation of evidence-based medicine guidelines. However, dissociation between what constitutes evidence-based medicine and the critical aspects of its delivery, noted by the plus (+) and arrow symbols, are the crucial process components necessary to implement any evidence-based practice.
Four aspects of Karl Weick’s proven model for successful emergent change were utilised in the early development of the project: specifically, aspects involving people in the process, providing clear direction, encouraging updating by acknowledging what is actually happening (as opposed to what might be or should be happening) and facilitating respectful interaction, which allows people to build a stable model of what they face were employed.
To that end, PDSA, a rapid cycle change model, and a well-defined and proven method for learning about and implementing change were used. This study sought to use the PDSA model in conjunction with ICU “disciplinary work.” We are defining this work as “disciplinary” because by comparison, there is much less in the way of discretionary decision-making with respect to which patients should and should not receive these interventions. Thus, system design should highlight the discretionary actions rather than obscure them by providing a disciplined means of ensuring appropriate care delivery and establishing the appropriate care action as the default.
We sought, then, to design a method of data collection that would reflect the process of care so that (1) system defects would be made apparent and (2) we could appropriately label certain actions as disciplinary or discretionary.
Forming a quality improvement work group
To investigate and define the scope of this problem, we formed a quality improvement work group for the DHMC ICU with the support of the ICU faculty leadership. The group consisted of two fellows, the ICU’s clinical nurse educator, two ICU research nurses, the ICU pharmacist and the ICU dietician.
The first task of the group was to define the steps in the implementation process, that is, the steps necessary to deliver the selected interventions to the patient (fig 2). We found that the process at the time was prone to possible complications by leaving crucial steps in the process open to the possibility of inaction.
The work group then designed a data-collection tool to highlight these process steps (table 1). Data were then collected in a prospective, “real-time” fashion on a group of 40 consecutive patients admitted to the ICU during April and May of 2004. We defined this as the control group. A change strategy was designed and implemented; the data were again collected on a second group of 40 consecutive patients and the results compared. Data on each patient were collected from admission until discharge or up to 7 days after admission, whichever came first.
The tool was pilot-tested for data integrity and inter-rater reliability. Our team used the data-collection sheet to collect two directly observed data points for each patient for each 24 h period. The two data points were designed to both add an element of stability to the data collection and to capture information from the two shifts, day and night, so as to involve the night-time staff, a group often ignored when examining care-delivery processes in the ICU. Additionally, the quality-improvement group met after a preliminary use of the data-collection sheet to further refine and create consistency with respect to the time and technique used when collecting the data.
Once the data were analysed, a change program was designed by the critical care work group based on the system defects demonstrated by the data for any measures that were less than 100%. The change strategy addressed three distinct areas: (1) team building, (2) system redesign and (3) enhanced communication. Based on the findings of Reasons,2829 we identified morning rounds as the optimal time for team-building in the ICU, because during this time every day all personnel involved in the care the patient were together. We revised the daily ICU progress note (fig 3), which essentially is the script for morning rounds, using an aviation industry preflight checklist as a model. Finally, we organised the sheet around organ systems with safety measures embedded in the assessment and plan as “go/no go” responses.
In the context of our change strategy, specific interventions were utilised to improve our main outcome measures of improving prophylaxis against venous thrombo-embolic disease, ventilator-associated pneumonia, and stress ulcers. To address team building, we conducted daily morning rounds. To address a uniform system redesign, we used a revised daily ICU progress note and an organ data sheet to prompt use of safety measures. To address enhanced communication, we maintained a real-time “data wall” that emphasised actual versus target implementation rates.
The revised daily ICU progress note was pilot-tested specifically to capture the processes of care as well as the implementation rates for the set of evidence-based measures, with an emphasis on simplifying the metrics. It contains prompts for the healthcare team to address each of the measures on a daily basis. The daily ICU progress note was integrated into the resident’s progress note that is completed and reviewed by the ICU team—consisting of an attending, a fellow, a resident, and the patient’s RN and respiratory therapist—for each patient every day.
During the change strategy phase of the project, the work group also created a “data wall” designed to provide the ICU staff with an ongoing, real-time image of our actual implementation rates versus our target implementation rates throughout the course of the project. The “data wall” consisted of a run chart with visual cues that indicated each stated goal and the position reached per week.
After previewing the revised resident goal sheet with the ICU team, the work group tested the process. During this phase, the measurement strategy incorporated the initial measurement tool as well as adding information on the proxy instrument—which in this case was the daily ICU resident sheet—to provide real-time feedback in data collection. The purpose of the proxy was to make the process of care transparent through live data collection used as part of the normal clinical rounding process, thereby reinforcing positive behaviour. Both traditional and non-traditional models were used to track real change and evaluate current performance. This information was used to continuously update and modify the delivery of care resulting in profound and statistically significant transformational improvements.
We performed a power analysis to ensure adequate sample size so as to avoid a Type 2 statistical error. Once collected and analysed, we hypothesised that these data would enable teams to target system defects in the delivery of care that resulted in implementation failures. Lastly, we calculated the implementation rates with respect to the identified measures.
Using the Indicated, Ordered, Received approach as outlined above, statistical process control charts were designed to examine two questions. First, if a specific metric was indicated, was it then received? Second, if a specific metric was ordered, was it then received? Data points were “batched” with respect to each measure so as to preserve time integrity and to ensure statistical stability. Additionally, data were collected to assess the ability of the “proxy”—that is, the daily ICU progress note—to reflect care as it actually happens in the ICU.
We recorded observations for each patient per 12 h nursing shift on three preventive measures. For any form of DVT prophylaxis, we recorded whether the sequential compression devices were turned on and were on the patient and/or whether the patient was receiving pharmacological prophylaxis (fig 4). For appropriate DVT prophylaxis, we separated patients by admitting diagnoses and then recorded how well their therapy correlated with the recommended therapy based on these diagnoses (fig 5). For each VAP prophylaxis ordered, we recorded whether there was an order placed for the head of the bed to be elevated to greater than 30° in patients when indicated (fig 6). Similarly, we recorded whether the head of the bed was raised greater than 30° in patients when indicated for every VAP prophylaxis performed (fig 7). Finally, for SUP, we recorded whether acid-suppressing medications were being given for patients on positive pressure ventilation and for patients with coagulopathy.
For each preventable measure, there was an increase in the measure of the mean as well as substantial narrowing of the control limits demonstrating improved process control.
We believe this project successfully provides new insights into improving patient care in healthcare settings. Most importantly, by the end of our study all metrics had increased significantly toward the goal of 100% success, thereby ensuring that evidence-based interventions improved care for ICU patients. Prior to our project, a hospital-wide quality assurance improvement initiative identified five preventable deaths over a 1-month period. Three of these deaths were directly related to a system failure with respect to patient safety: Two were the result of a lack of DVT prophylaxis and one the result of a lack of stress ulcer prophylaxis. Our study interventions substantially increase the likelihood that these safety measures would be effectively integrated into the system of care.
Additionally, unlike currently available methods that define only the delivery of the appropriate intervention as success, our unique approach offers the opportunity to capture true negatives as well as true positives to offer a more comprehensive system review. While previous studies have rightly emphasised delivering care when it is indicated,19–22 our study sought to incorporate measures to ensure that care is not delivered when none is indicated so that patients do not receive unnecessary and potentially harmful treatments. These interventions provide an excellent teaching resource, as they reinforce the application of appropriate and timely care. The method of implementation enhances team communication and interaction focused around the progressive needs of the patient.
Beyond the implications of our study for improving the process of ICU care, we also developed a methodology that fine-tuned the PDSA process itself. For example, our method of data collection allowed simplified defect analysis, thereby eliminating resource-consuming audit of data after the fact. On a larger level, through this study, we believe we have developed a durable and reproducible intervention model that can be applied to various aspects of any healthcare microsystem that is process-oriented.
The basic approach we developed and followed, while specifically designed for an ICU setting, provides other settings with an adaptable process that can be tailored for use in other healthcare microsystems. The interventions presented in this study, therefore, are not meant to be universally applicable but are intended to provide a framework for initiating team building and evidence-based care.
Finally, this work suggests an opportunity to develop a critical care patient bundle in that it allows for: the incorporation of smaller, more discrete bundles of work such the “ventilator bundle,” allows for the inclusion of all patients admitted to a critical care setting as half of all patients in critical care units are not ventilated and yet still require several elements of disciplinary care, and allows for a process of continually updating as new information and new guidelines become available and require implementation.
Funding: Supported by a grant from the Quality Research Grant Program.
Competing interests: None.