Embedding Two of the 4Ms Into the Electronic Health Record: One Health System’s Journey Towards Systemwide Spread of Age-Friendly Care

Abstract

The purpose of this article is to describe 2 quality improvement projects aimed at embedding 2 of the 4Ms into the electronic health record for system-wide spread of Age-Friendly care. The 2 projects described in this case study serve as exemplars for the future implementation and sustainability of 4Ms care. Rapid-cycle quality improvement projects, via the Plan, Do, Study Act model, focused on the 4Ms were conducted by interprofessional teams to integrate clinical decision support for clinicians within the electronic health record. Project Senior Care Review for Evaluating and Eliminating Non-essential and potentially inappropriate medications (SCREEN) embedded a geriatric medication screen into the ordering panels of the top medications identified as being prescribed to older patients potentially inappropriately. Project Predictive Real-time Evaluation of Delirium in Clinical Therapy (PREDICT) embedded a delirium prediction rule in the electronic health records to guide clinicians to implement delirium mitigation interventions on patients at risk of developing or experiencing delirium. Outcomes were evaluated descriptively utilizing data and reports generated by the electronic health record. Embedding non-interruptive and actionable clinical decision support in the electronic health record supported the rapid spread of Age-Friendly care across a 7-hospital system. The 4Ms can be embedded into existing workflows through novel implementation of best practices by leveraging the electronic health record. By embedding 2 of the 4Ms into existing workflows and creating non-disruptive, actionable clinical decision support within the electronic health record, clinicians have the tools to implement Age-Friendly care within the 4Ms framework. Additional projects aimed at embedding the other Ms are underway, and long-term outcomes are being evaluated.

Highlights

● Two clinical decision support tools embedded in the electronic health record to support age-friendly care across a health system.

● Leveraging rapid-cycle quality improvement processes can faciliate the implementation of age-friendly care.

Introduction

In 2017 Hartford HealthCare began its Age-Friendly journey by identifying components of 4Ms care already implemented in clinical practice that supported positive patient outcomes. Geriatric-focused services were present throughout the healthcare system, consisting of 7 hospitals, 3 skilled nursing facilities, 2 assisted living facilities, 8 outpatient geriatric primary care or specialty clinics and a home health care agency. Within these sites, several programs stood out as models demonstrating successful Age-Friendly care. These included an inpatient delirium care pathway-Actions for Delirium Assessment Prevention & Treatment (ADAPT),^1,2 an outpatient geriatric oncology program- COACH ³ - and a home based APRN visiting program. As the growth of the healthcare system outpaced the growth of the geriatric program and geriatric-certified team, it became evident that all clinicians would need support in caring for older adults across the care continuum. With the implementation of a system-wide electronic health record (EHR), Epic, the most efficient way to implement Age-Friendly care was to embed best-practices into EHR functionality.

The literature supports numerous ways to deliver Age-Friendly care utilizing the EHR and 1 way to address the Medication “M” is through passive medication clinical decision support for geriatric dosing of high-risk medications. Several studies have demonstrated a positive impact of implementing computerized physician order entry (CPOE) defaults as a feasible and sustainable way to improve medication safety for older adults.^{4 -7} Health care providers are challenged to select the optimal type, dose, frequency and combination of medications for medical treatment while simultaneously avoiding unintended deliriogenic effects of medications. A systematic review and meta-analysis found computerized interventions effective in reducing potentially inappropriate medications (PIMs). ⁷ These include clinical decision support (CDS) systems and alert systems, that provide guidance to prescribers. The CDS approach can include CPOE which is more proactive, defaulting to desirable dosages and frequencies as a medication order is placed. The alert system is more reactive, firing a warning after a potentially inappropriate or high-risk medication order is entered. Although both approaches have been shown to be successful in reducing PIMS, the default doses are often higher than recommended for older adults ⁸ and the alerts often cause alert fatigue ⁷ leading to resistance from prescribers to accept the recommendations.^9,10 One study found engaging prescribers in the design and development of the intervention improved compliance. ⁵

Similarly, the EHR can be leveraged for predicting delirium risk for hospitalized patients based upon risk factors automatically pulled from patient data and calculating a risk score.^{11 -13} Many studies have aimed to identify risk factors that contribute to delirium.^11,14,15 This has generated an extensive list of both pre-existing and precipitating risk factors for clinicians to consider when trying to prevent delirium. Unfortunately, this has limited clinicians’ ability to identify a high-risk population to target interventions, as so many patients have 1 or more of these risk factors. A variety of predictive models have been presented in the literature to streamline this list by prioritizing selected risk factors.^12,13 A systematic review of prediction models found a lack of consistency and limited predictive ability among existing models. ¹⁶ Even though no one tool has been shown to be optimal, there are several risk factors that appear in most of the models. ¹⁵ By having these prioritized risk factors automatically pulled from patient data and calculating a risk score, clinicians can be guided as to the population toward which to target delirium prevention and detection efforts. ¹⁷

By implementing these tools into the system-wide EHR, the 4Ms can be delivered across care settings- even when geriatric consultative services are not readily available. The aim of this case study is to describe 2 quality improvement projects that embedded 2 of the 4Ms into the EHR for system-wide spread of Age-Friendly care.

Methods

Project Senior Care Review for Evaluating and Eliminating Non-essential and Potentially Inappropriate Medications (SCREEN)

Plan

Project SCREEN involved an interprofessional workgroup consisting of representatives from nursing, providers, pharmacy and informatics. This workgroup met consistently for a year to review options for clinical decision support for clinicians around medications that contribute to patients’ risk of falling and delirium. The team considered several iterations of alerts, banners and other types of clinical decision support before deciding on a geriatric clinical context screen, largely due to the success of a similar intervention by Mozer et al., 2022 ⁶ in changing provider prescribing patterns and provider preference to avoid disruptive alerts. The decision was made to build in non-interruptive clinical decision support and add a safety banner to selected medication order panels to provide the most efficient workflow for providers. This geriatric clinical context screen would apply to the order panel of select medications for any patient age 65 and older.

The screen would display suggested geriatric-safe doses and frequencies for the medication as well as a banner in the order panel indicating that this medication increases the patient’s risk of falls and delirium. The first set of medications selected for this project were based on medications most frequently ordered for patients 65 years and older in potentially inappropriate doses. This was determined by the informaticist pharmacists running reports on Beers List ¹⁸ medications ordered for patients 65 and older. Of the 27 medications on the Beers List with the most orders in the EHR for patients 65 and older in the previous 6 months, the medications with the highest total orders or the with most clinical implications based on expert review by the interprofessional workgroup were selected for the first round of the geriatric clinical context screen. Total orders ranged from 97 (diphenhydramine) to 12 202 (lorazepam injection). Nine discrete medications were chosen for the first round of the geriatric clinical context screen application. The decision was made to implement the screen in waves instead of applying it to all 27 medications that were potentially inappropriate for older adults because the screen required extensive EHR build time from the informaticist pharmacist and the workgroup wanted to see outcome data indicating the screen’s positive impact on reducing potentially inappropriate medications before committing the resources to applying it to all of the Beers List medications.

The interprofessional workgroup reviewed the medications and made recommendations for appropriate doses and frequencies based upon expert opinion, recent literature and recommendations from expert groups such as the American Geriatrics Society. ¹⁸ Once the medications and suggested doses and frequencies were chosen, the proposed geriatric clinical context screen build was presented and approved by several system-wide councils with stakeholders across the care continuum as any changes within the EHR are applied system-wide.

Do

The project went live in November 2023. The geriatric clinical context screen was applied to all 9 of the selected medications for patients 65 and older including inpatients and patients in the Emergency Department. Medications within pre-built ordersets, such as post-operative pain ordersets or delirium prevention ordersets that had pre-existing medication orders, were excluded from the screen. See Figure 1 for an example order panel comparing the pre-intervention order to the post-intervention order and noting that in the post-intervention order only the geriatric specific doses and frequencies are visible.

Figure 1.

Change in order frequencies of selected medications 90 days post implementation of Project SCREEN and example order panel changes.

Study

Six months post-implementation the workgroup re-convened to review any feedback from end-users and to review adherence to the screen’s recommendations. There were some concerns related to the suggested dose for haloperidol specifically in the critical care setting. This allowed the group to review literature specific to haloperidol dosing in critically ill patients and provide an opportunity to leverage the existing Intensive Care Unit (ICU) delirium orderset to treat delirium in critical care settings. The ICU delirium orderset contains higher doses of medications, including haloperidol, that would be more appropriate for that patient population.

In addition to the anecdotal feedback, Slicer Dicer data models were run in Epic to review adherence to the geriatric screen recommendations. Slicer Dicer is a data exploration tool in Epic, that allows the end user to mine data from a large population of patient records within the electronic health record. ¹⁹ Results from the 90 days pre-intervention and 90 days post-intervention are presented in Figure 1 which demonstrate that every medication included in Project SCREEN exhibited a change in prescribing pattern toward the recommended dose except for morphine. Every medication included in Project SCREEN exhibited a change in prescribing pattern toward the recommended frequency except for diazepam, morphine and quetiapine. However, the difference in percentage of orders consistent with the recommended geriatric dose (t(24) = −0.51, P = .30) or frequency (t(24) = −0.03, P = .48) pre-intervention compared post-intervention was not statistically significant. During the Leapfrog Computerized Physician Order Entry (CPOE) testing survey, which ensures that at least 75% of medication orders are entered via a computer system with software to reduce prescription errors, Hartford Hospital was tested in the “inappropriate based on age” category of CPOE testing. The geriatric clinical context screen satisfied the requirements from Leapfrog by flagging the high-risk medication based upon the patient’s age, and also guided the clinician to the appropriate dose which translated to a better clinical outcome without increasing alert fatigue.

Act

Six months after Project SCREEN went live, the interprofessional team met with stakeholders from the ICU and made Epic enhancements so that when providers searched for haloperidol in the order-entry field in the EHR, the associated ordersets containing haloperidol, such as the ICU and non-ICU delirium ordersets- would appear for selection. Based upon the positive performance in the Leapfrog CPOE testing, the Pharmacy and Epic teams were enthusiastic to expand the geriatric clinical context sceen to additional medications. The interprofessional workgroup began to meet biweekly to select the next set of medications to apply the screen to and decide on suggested doses and frequencies.

Project Predictive Real-Time Evaluation of Delirium in Clinical Therapy (PREDICT)

Plan

Based on findings from an ongoing quality improvement project aimed at proactively addressing escalating patient behaviors, an opportunity existed to intervene earlier for patients at risk of or experiencing delirium. The ongoing project consisted of a Behavioral Time Out that allowed any member of the healthcare team to call together members of the care team together in real time to address patient behaviors such as attempts to get out of bed, behaviors requiring a one-to-one sitter or the use of restraints, patients attempting to leave against medical advice or escalating alcohol withdrawal symptoms. Based on preliminary, unpublished findings from this project, the primary opportunity was the assessment of and proactive management and prevention of delirium. The opportunity to intervene and prevent the development of delirium exists within the first 48 h of admission based upon a review of several million Confusion Assessment Method (CAM) ²⁰ assessments from Hartford Hospital, noting that of patients who were ever assessed as CAM positive- delirious- during their hospital stay, 68% were assessed to be delirious within 48 h of admission. Consulting geriatrics or putting delirium mitigation interventions in place after a patient was already CAM positive was already too late for most patients- preventing the development of delirium would have a greater impact on patient outcomes.

An interprofessional stakeholder workgroup consisting of representatives from nursing, geriatrics and clinical informatics gathered to assess different workflow options within the EHR to best provide clinical decision support for clinicians to prevent delirium for hospitalized patients. Prior to this project, the nurses received an interruptive alert- or Best Practice Advisory (BPA) ²¹ - to notify providers of a patient’s first CAM positive assessment. BPAs provide clinical decision support in the form of an interruptive alert to the clinician with display text containing a warning or reminder along with follow-up suggestions that can be selected. ²¹ However, anecdotal feedback and an analysis of provider workflows after this notification, if it occurred, was that the providers did not understand what a positive CAM assessment indicated nor what the next steps for clinical treatment should be.

The interprofessional Project PREDICT workgroup examined other predictive risk models within the EHR. One existing risk model utilized on the maternity units was the “infant drop risk” prediction rule that pulled risk factors from the birthing parent’s chart to predict if the birthing parent was at risk of dropping their infant during their hospital stay and displayed a BPA to the clinician if a combination of those risk factors were detected in the chart. The Project PREDICT workgroup mirrored the delirium risk prediction rule after the infant drop risk prediction rule. The intent of the delirium prediction rule was to capture the risk factors within the EHR placing that patient at risk for developing or experiencing delirium and display a BPA to the clinician with recommended actions. The interprofessional team reviewed the literature citing similar predictive models and selected risk factors based on several studies^{11 -13} and that were discrete fields captured on most patients within our institution’s EHR. See Table 1 for the risk factors chosen for the delirium prediction model, the threshold value needed to have the risk factor weigh into the total score to trigger the BPA and the location within the EHR that the risk factor can be found. Once the risk factors were chosen for the predictive model, they were assigned a weighted score and a threshold score of 3 or more was chosen to trigger the clinical decision support. See Table 1 for the weights assigned to each risk factor. The weights were chosen by the workgroup based upon evidence identifying it as a risk factor for delirium and the desired sensitivity of the delirium prediction rule by the workgroup. For example, a positive CAM score was assigned a weight of 3 as the workgroup wanted the delirium prediction rule BPAs to fire for any CAM positive patient so that the clinician would see the BPA and take the recommended actions.

Table 1.

Delirium Prediction Rule Risk Factors and Their Locations Within the EHR. The Points Assigned Represent the Weight of Each Risk Factor in Calculating the Total Score to Trigger the Delirium Prediction Rule Best Practice Advisory to Fire for the End-User. The Value Represents the Threshold of the Risk Factor That is Needed to Have the Risk Factor Weigh Into the Total Score. The Threshold was Increased to 5 or More Points 2 months Post Go-Live.

Risk Factors	Points	Epic Location	Value to trigger score
Age ≥ 65 years old	1	Age	≥65 years old
Baseline mental status	2	Patient Profile: Baseline Mental Status field	≠Within Defined Limits
Active delirium	3 Upated to 5 two months post go-live	Adult PCS flowsheet rows: CAM or CAM-ICU	=Positive
History of delirium/encephalopathy/ dementia/neurcognitive disorder	2	Problem List or History	ICD10 Codes present
Hearing or visual impairment	1	Adult PCS flowsheet row: Fall Risk Screening- Significan Visual or Hearing Impairment	=Yes
Dehydration	1	Lab Results: BUN/Cr ratio	>10
Hip fracture	1	Admitting diagnosis of hip fracture	ICD10 Codes present
Total score for delirium prediction rule BPA to fire	≥3 Updated to ≥5 two months post go-live

The workgroup designed clinical decision support for both providers and nurses consisting of actionable BPAs. The BPA for the nurses guided them to add the evidence-based delirium risk clinical practice guidelines which also added patient and family education and flowsheet intervention rows providing delirium prevention and treatment interventions. The provider BPA was also designed to notify the providers that their patient was identified as being at risk for delirium and guide the providers to order the appropriate delirium orderset, ICU or non-ICU, depending on their clinical setting. The delirium prediction rule build and BPAs were reviewed by several system-wide interprofessional councils to solicit feedback and support.

Do

The delirium prediction rule was launched in March 2024 without the BPA’s firing to the clinicians to ensure the delirium prediction rule was firing on appropriate patients as well as to assess the potential BPA burden to clinicians. The BPAs went live to clinicians in September of 2024.

Study

Between March and September of 2024, the Project PREDICT workgroup performed weekly reviews of the patients that the prediction rule was firing for and using expert review to determine whether the prediction rule was firing appropriately. With the partnership of our clinical informatics team, the Project PREDICT workgroup also reviewed the number of BPAs that would be firing to providers and nurses to understand the potential alert burden for end-users.

In addition to the weekly workgroup reviews of Project PREDICT’s performance, a quality improvement project aimed at increasing geriatric consults on a 46-bed medical-surgical unit, the largest unit at the system’s largest and flagship hospital, was conducted in parallel to the launch of Project PREDICT. Patients 65 and older admitted to the medical-surgical unit that triggered the delirium prediction rule were reviewed for documentation of a positive CAM assessment and a consult to geriatrics was placed when appropriate. This systematic process of reviewing patients triggering the delirium prediction rule served as the first step of validating that the predictive rule was identifying appropriate patients based on expert review by the quality improvement project team. Feedback from the quality improvement team providers that were actively caring for the patients identified by the delirium prediction rule was provided to the Project PREDICT team daily to gain valuable insights into the accuracy of the prediction rule and as a first step in evaluating the sensitivity and validity of the tool. Further sensitivity and validity testing will need to be completed prior to dissemination of the predictive tool. A formal evaluation of the concurrent quality improvement project with clinical outcomes associated with early geriatric consultation for patients identified as being at risk for delirium is pending and will be described in further detail once the pilot is completed.

Once the BPAs went live to end-users, all BPA feedback was reviewed by the Project PREDICT workgroup and Slicer Dicer models were run to analyze frequencies of the BPAs as well as conversion of the BPA to ordering the delirium orderset. End-users, providers and nurses, can provide feedback within Epic when the BPA fires regarding the appropriateness and utility of the BPA. This feedback, which consists of frequencies of “likes,” “dislikes” and free-text comments were reviewed by the clinical informatics team and then reported to the Project PREDICT workgroup to guide any potential changes needed for the prediction rule. Providers indicated that they needed a way to bypass the BPA besides ordering the delirium orderset or selecting “Remind Me Later” which only prevented the firing of the BPA for 15 min. Some providers noted that their patients were end-of-life and that delirium prevention and treatment strategies were not necessary. Slicer Dicer data models were run in Epic to descriptively analyze frequencies of the BPAs and conversions to delirium orderset orders. Figure 2 depicts the frequencies of the actions taken by the provider when presented with the BPA. Out of 3994 patients for which the BPA fired, the delirium orderset was opened and added to the provider order panel for 1353 patients, representing the 34% of patients identified as being at risk for delirium for which the provider chose to order the delirium orderset. Figure 3 indicates a statistically significant increase in providers ordering the delirium orderset in the 6 months after project PREDICT launch compared to the 6 months before project PREDICT launch (t(10) = −5.06, P = <.001).

Figure 2.

Epic SlicerDicer Data Model results for delirium prediction rule provider BPA 1-month postimplementation. These results depict the frequencies of the actions taken by the provider when presented with the BPA. Out of 3994 patients for which the BPA fired during 1-month post-go live, the delirium orderset was opened and added to the provider order panel for 1353 patients.

Figure 3.

Number of Delirium ordersets ordered by providers 6 months pre/post implementation of Project PREDICT.

Act

Based upon review of the BPA feedback, frequency of BPA firing and the conversion rate of the BPA to delirium orderset utilization, the interprofessional workgroup re-convened 1 month after the BPAs went live and reassessed the prediction rule risk factors, the patient-level criteria applied to the rule and the options for end users to act upon the BPA. Adjustments were made to the patient-level inclusion criteria- mainly excluding patients that were end-of-life. The trigger threshold score was also increased to 5 or more with the CAM positive risk factor weight increasing to 5. This increase in total threshold score is indicated on Figure 3 at the start of the month of November. Additional selections were created for providers to prevent the BPA from firing for 24 h and for the duration of the patient’s admission. The team reconvened monthly to review feedback and consider any revisions to the predictive model to make it as clinically significant as possible.

Discussion

Interprofessional engagement and collaboration proved fundamental to the integration of Age-Friendly care into the EHR. The engagement and support from our pharmacy, clinical informatics and Epic teams were critical to the success and expansion of Project SCREEN. A strategic stepwise approach to implementation, by choosing to include only a few medications at a time, enhanced acceptance and adoption so that the team did not feel overburdened by the projects’ aims. The team also solicited feedback from front line staff via clinical councils and project team meetings between progressive builds to make course corrections and enhance ownership from end-users. Formal interviews, focus groups or qualitative assessments were not implemented to collect this feedback or measure the extent of the interprofessional collaboration of the project workgroups during the PDSA cycles, however this is an opportunity for further exploration.

Project PREDICT was spurred from requests made by clinical staff to be made aware of patients at risk of experiencing delirium at the time of hospital admission or when a change in condition occurred. Hartford Hospital’s robust Actions for Delirium Assessment Prevention & Treatment (ADAPT) care pathway incorporates delirium prevention measures and a delirium screening mechanism. ²² ADAPT indicates risk factors for delirium but does not alert the front-line staff that these are present, therefore potentially missing an opportunity for staff to implement proactive and preventative interventions. Recognizing that early detection and intervention is paramount, the implementation of the delirium prediction rule in Emergency Department settings with the creation of Emergency Department delirium ordersets is a necessary next step.

Hartford HealthCare has several workflows for the assessment and intervention of the other 2 Ms- Mobility and What Matters Most. However, non-interruptive clinical decision support is not utilized, therefore creating an opportunity for improvement in the delivery of the 4Ms as a set.

Integrating 4Ms workflows is one of the first steps toward the spread of Age-Friendly care, however clinicians need tangible tools and resources to manage older hospitalized patients. To implement several of the delirium prevention strategies indicated in Hartford HealthCare’s delirium ordersets, therapeutic and pharmacologic items need to be readily accessible to front line staff. Hartford Hospital has leveraged the use of trained volunteers to administer or facilitate the implementation of preventative interventions, stocking therapeutic items locally on patient units and putting processes in place for proactive rounding and restocking. ²³

In a recent scoping review, Schöler et al. ²⁴ examined 120 studies describing the development of delirium prediction models using data derived from the EHR and found that only 1.7% of the models were actually implemented in the hospital setting. This underpins the primary concern with the development of sophisticated prediction models- they are not consistently implemented in the clinical environment to guide interventions for patients at risk for or experiencing delirium. Project PREDICT represents an evidence-based predictive model that was implemented in rapid PDSA cycles based on available data rules already built into the EHR. Through clinical implementation, the model is actively improving care for high-risk patients and real-time clinical feedback is used to adjust the model to make improvements in sensitivity and for validation of the model.

The impact of Projects SCREEN and PREDICT has not been fully evaluated. Clinical outcomes such as adverse medication events, delirium rates, delirium-attributable days, falls, restraint use, one-to-one sitter utilization and length of stay have not been associated with the project implementations, but will be evaluated in future analyses. As Figure 2 indicates, the majority of patients for which the delirium prediction rule BPA fired, 66%, did not have a provider order the delirium orderset. This requires further exploration by the interprofessional workgroup into the sensitivity of the delirium prediction rule and its ability to capture appropriate patients at risk for or experiencing delirium. A limitation of these rapid PDSA cycles is the lack of robust statistical analysis to rigorously assess the success of these clinical decision support tools. Future evaluations and research will include additional analyses to provide a more comprehensive understanding of the impact of these projects on long-term patient outcomes, clinician prescribing and ordering patterns and cost-savings to the organization.

Conclusion

Embedding Age-Friendly workflows into the EHR is paramount to the implementation and spread of the 4Ms. Clinicians increasingly rely on the EHR for guidance and clinical decision support to provide the most evidence-based care to patients- particularly older adults with multiple chronic conditions and increasing complexity of care needs. Reducing clinical cognitive burden by minimizing disruptive BPAs is essential to promote efficiency and safety within the acute healthcare environment. Similar models for the other Ms- Mobility and What Matters Most- could be beneficial for the consistent and standard delivery of all 4Ms.