VisualMED Clinical Solutions Corporation

Electronic Health Records and Quality of Care

 
Synopsis of a Presentation delivered at the Clinical Automation Summit
by Sam Huntley, RN
September 17, 2007
Chicago, Illinois
 

Good afternoon

My name is Sam Huntley and I’m a Nursing Administrator at a small hospital in West Texas where I serve in rotation as Administrator on Duty with other senior nursing staff. Prior to joining the hospital where I am currently located, I worked for 17 years in Critical Care nursing at a nearby facility, where I was involved in assessing new technology for the critical care environment. About a week ago, I was asked if I could come up to Chicago to present data drawn from the hospital’s experience with its EHR and decision support system, acquired from VisualMED Clinical Solutions Corp back in 2004.

When I looked up the brochure for the meeting over the internet, I saw that the material to be covered in this session was originally centered around the EHR experience of an unnamed Phoenix hospital, who, I assume were selected as an example of an organization which implemented a clinical information system as part of a process to foster best practice and achieve fully electronic patient care documentation. They sounded like the type of organization well-positioned to analyze the impact of an EHR on clinical outcomes and other indicators. They certainly sounded like the type of organization that had 100% staff buy-in with respect to the implemented EHR, with widespread support for the development and evolution of the system into which they had invested time, resources, and money.

There’s nothing at all I can tell you about that Phoenix hospital. But what I’m going to tell you about, is a small hospital in the real world, a hospital that doesn’t run 90% occupancy, that didn’t have a master plan, whose administration never committed to implement paper-free documentation, and which has undergone major shifts in physician and investor support in the three years of its existence. The hospital I’m going to tell you about is a about as far away from a model EHR site as you can get. From a certain perspective however, these conditions may make the story a more interesting one.  

Built in 2002 and 2003, the hospital opened its doors in January of 2004 as a for-profit institution with about 30% physician ownership. Currently, the hospital runs 40 inpatient beds, and maintains a census of between 60 and 75% full. There are 6 operating rooms, 2 endoscopy suites, and a cardiac catheterization lab.

About 200 physicians maintain privileges at the hospital, with about 120 core physician staff covering a wide variety of specialties.

The hospital informatics “backend” is handled by Meditech components for ADT, laboratory, radiology, pharmacy, and pathology. A Medquist dictation system is available, as is a PACS and advanced medical imaging systems. A conscious effort was made by the founding administrative staff to find a clinical system front end through which all staff would access healthcare informatics information, rather than using Meditech for this purpose. This was because the senior administrators, including the CEO at the time, came from clinical backgrounds, came from Meditech backgrounds themselves, and were looking for something that addressed clinical issues from a clinical perspective. The VisualMED Clinical System was selected for that purpose.

The clinical information system with its CPOE, decision support, and EHR components was installed in the Spring of 2004, went live in June of that year, and has been in operation continuously since, for clinical care documentation of all inpatient activities, endoscopy, cardiac catheterization, pre-op, post-op, and E.R. The system communicates with the appropriate backend Meditech systems, including the pharmacy, and is able to integrate laboratory results reporting, allergies, and findings documented in the EHR with decision support. 

So, how can an EHR have a measurable impact on the quality of care?
Should we expect a decreased length of stay for defined DRG’s?
Should we expect a decrease in the number of adverse drug events?
Should we expect decreased morbidity and mortality related to those avoided ADE’s?

The literature responds to all of these with a “YES.” I believe a thorough statistical analysis of all aspects of care at the hospital would demonstrate them as well. But I don’t have any of those numbers to share with you this afternoon. I would instead like to go through some smaller-scale scenarios that nevertheless document the measurable impact that an EHR really has on quality of care.

These positive impacts are felt where the EHR helps the hospital to meet regulatory requirements, where it helps the hospital to meet internal objectives, and where it supports best practice.  

The JCAHO “Do not use” list was developed in 2003 and 2004 and ratified at the November 2004 National Summit on Medical Abbreviations. As shown in this slide, the list contains a number of Latin-derived and other entries that came to the attention of the Joint Commission following their repeated implication in medical prescription error in audits carried on in 2001 and 2002. Though designated for use in paper-based documentation or in electronic free-text documentation, the hospital requested the implementation of the “rationalized” nomenclature during a JCAHO site visit that took place 2 weeks after the clinical information system was implemented. The requested changes were made to the EHR e-prescription content during the afternoon of the first day of the JCAHO site visit. By the time the inspector returned the following morning to complete her evaluation, the “Do not use” list had been implemented in its entirety, including several of the “optional” items on the list. As a result, the hospital achieved 100% compliance in this important area in 2004, reducing iatrogenic risk clearly documented in the JCAHO reports of the day. Of note, the JCAHO website today reports that compliance with the requirements of the “Do not use” list reached 78% of JCAHO-affiliated sites only in 2006, the most recent year for which data is available.  

Use of the EHR allows the hospital to evaluate itself according to the core competencies are established from time to time by JCAHO and other regulatory organizations.

A recent core measure assessment involves the determination by a site that all patients treated for heart failure as an inpatient be discharged with a prescription for either an ACE inhibitor or an ARB unless specified contraindications to such medications are noted in the chart. Such contraindications might include a history of intractable cough due to the medication, high serum potassium, or a moderate degree of renal failure.

The graph in this slide shows the percentage of all patients with an admitting diagnosis of heart failure, as specified by ICD9 code, who at discharge were given a prescription for a drug in the class of either ace inhibitors or ARB. The graph covers a 3 year period, beginning at the time of the implementation of the clinical information system. In the first year, about 55% of patients with the index diagnosis received the target medication, in the second year, about 64% of patients, and in the third year, which ended just this past July, about 85%. The total number of patients over this period was just 75, so it is very possible that true statistical significance is not attained. There is nevertheless a very real trend and if we look more closely at the data, the results get even better. 

This slide shows the same data as is displayed in the preceding slide. However, here we have taken a look at those patients who had the index diagnosis of heart failure, but who did not receive either of the target medications, AND who were found to have a high serum potassium. A high serum potassium is one of the relative contraindications to prescription of the target medication according to the JCAHO fact sheet which accompanies the published core measure requirement. Patients who had the index diagnosis, were given no target medication and who had a high potassium are shown in orange in the bar graph. The y axis is again expressed in terms of percent. Each blue and orange bar therefore represents the sum of those patients treated correctly plus those not treated, but also, correctly. In other words, the physician made a correct intervention by not treating a given patient if that patient had a contraindication to treatment. Looking at the numbers, in year 1, 65% of index patients were treated correctly; in year 2, 70%, and in year 3, nearly 90%. In fact, it’s possible that the clinical care was even better, as this analysis doesn’t take into account those patients who might have had intractable cough due to the administration of one of the target medications, nor those patients with more than moderate renal failure. Given the time available to put together this presentation, the analysis was not carried further. Be that as it may, the clinical staff maintained a very respectable record on their own, and that record improved with time.

Individual hospitals may have internal requirements independent of those mandated by regulatory organizations. A requirement for cost-containment is one of these, and one which is well suited to implementation through an EHR system.

Our hospital recently took on a new Pharmacy Director who reviewed drug usage history and developed a plan to rationalize drug use at the site. Prior to the start of this process, any staff physician could prescribe any medication, and that medication would be obtained from an off-site pharmacy even if a generic equivalent existed at the hospital. As you can imagine, the pharmacy budget required to maintain such a habit was not sustainable.

One of the medications targeted by the pharmacist was pantoprazole, used to treat gastric hyperacidity. The very small pharmacy staff was not in a position to seek out every physician following a pantoprazole prescription, nor were they in a position to issue an ultimatum to physicians, who, in El Paso, have a reputation of being somewhat unruly. The solution seemed to be to build a drug warning algorithm inside the EHR decision support module that would remind the prescriber that an equivalent cost-efficient alternative was available, and that it was in everyone’s interest to use it.

This slide shows a portion of an EHR administrative decision support screen showing how the message to the prescriber, specified by the pharmacist, would be linked to the system recommendation for an alternative proton pump inhibitor, omeprazole. The per tablet cost to the hospital for the pantoprazole was $4.50 while the cost to the hospital for the alternative omeprazole was just 60 cents.

Below the screen sample, we see the data for the first few months of operation of this drug warning, normalized to one full year. In the year ending immediately prior to the initiation of the cost containment program, expenditure in the pharmacy budget for pantoprazole came to $42,637. Normalizing data to one year, based on data collected to date, pantoprazole prescriptions are expected to fall by 41%, in view of the fact that in a little more than 3 months, the algorithm as been triggered 138 times, and in 57 instances the prescriber either cancelled the pantoprazole prescription altogether, or accepted the system-generated alternative, omeprazole. We can therefore anticipate that the annual expenditure for pantoprazole will drop to about $25,000 and though there will be an increase in expenditure for omeprazole, that increase will amount to just $1,446. Taking into account that the prescribers cancelled about 15% of their pantoprazole prescriptions without substituting any alternative, there is a net saving to the hospital of $16,165 in the first year. In fact it is likely that savings will increase in subsequent years as repeated exposure to the algorithm trigger causes more resistant staff to change their prescribing habits. Review of the medical literature shows that exposing prescribers to the actual cost of the prescribed medication results in reduced prescribing costs over time.        

Essentially every hospital has a requirement for the internal publication of a monthly infection control report. How many hospitals have a requirement for a monthly infection control report that is actually produced within a day or two of the end of every period? A timely infection control report can be used as a surveillance instrument, a much more powerful tool than a simple reporting document.

The EHR is used at the hospital as the basis of the monthly infection control report. Infection control staff routinely document infection control issues in each patient’s EHR. An informatics query is then applied to each EHR at the end of each reporting period. Content of the query is described by the list of field names typically accessed from the EHR database table, shown in the center of the slide.

An excerpt from a paragraph-format sample of a typical infection control report is shown at the bottom of the slide. Though not shown in this particular example, the EHR query takes patient room numbers into account so that disease spread may be examined should clusters of MRSA or other pathogens be noted in one or another hospital location.

Introduction of EHR is a real impetus to best practice, and I’d like to present some evidence from our hospital in support of that claim. But first, a bit of background: There are about 200 physicians on staff at the hospital, many of whom work at multiple facilities nearby, in some cases seeing patients at as many as four other hospitals in the course of a day. The actual physician presence in-hospital varies widely, as does the closeness of any given staffer to his or her case. Nurses take the real responsibility for continuity of care for many patients, and are expected to communicate with off-site doctors as they deem appropriate. In this setting, therefore, the responsibility for prescription order entry is divided among nursing and physician staff, although all orders are initiated by the physician. The patient mix is as one might expect in a community hospital, and some patients are quite sick. There are a small number of respirator dependent patients, and some receiving TPN at any given time. In a poly-pharmacy environment, nephrotoxic medications are prescribed routinely. The questions are these: (1), Does the hospital adhere to best practice in regards to caring for its patients with kidney failure? and (2), Does an EHR with integrated decision support make any material difference to practice?

Developing the answers to these two questions will occupy the rest of our session. The type of analysis we will review is not unique to our hospital but applies to any order entry pathway for any medication with narrow risk/benefit ratio and where optimal dosing can be defined and predicted by combinations of laboratory, demographic, and physical findings.

This slide depicts the segment of a screen in our EHR system that the System Administrator uses to set internal criteria for dose modifications for any drug in the hospital formulary, all of which are prescribed through the EHR decision support module. For the purposes of discussion, I’ve highlighted the row labeled gentamicin at the very bottom of the slide. The selection of that row causes the renal failure dose modification method to be displayed for that drug in the upper portion of the screen fragment. The method is displayed in the lavender bands in the middle of the screen. The EHR formulary is a mirror of the hospital formulary which takes into account clinical business rules rather than inventory parameters and contains about 3000 listings, any one of which may be associated with a renal dose modification rule. These rules are all empiric, are derived from the medical literature, and were developed, tested, and validated by the vendor’s pharmacy staff. They conform to methods routinely used for renal dose modification at any clinical center. The rules make use of the basic information any doctor should have at his or her fingertips when prescribing nephrotoxic drugs: simply, the patient’s age, sex, height, weight and a single laboratory test result, the serum creatinine. This then is the architectural framework for the decision support we are about to review.

This slide shows a portion of the EHR decision support screen that is displayed to the prescriber when he or she has ordered a nephrotoxic medication. The system has already completed its real-time analysis of the order and determined that a dose modification is necessary. The beige band at the top of the image displays the text of the gentamicin prescription as specified by the prescriber. The pink band below that displays a modified version of the order that conforms to best practice standards given the patient’s degree of renal failure. In order to prescribe the correct dose of the medication, all the prescriber needs to do is to press the button labeled “Save.” In order to prescribe what is essentially the wrong dose, the user must modify the dose amount highlighted in dark blue in the example or modify the dose frequency using the screen control to the right, and then press “Save.” The bottom line – the prescriber has to go out of his or her way to prescribe the wrong dose, rather than press a single key to prescribe the correct dose. And this is not a question of local practice style, nor one of a choice among a set of safe, equivalent doses. There is only one right answer. Every doctor knows this. Nearly every doctor knows how to do the calculation, and if not, can get the answer from the hospital pharmacist over the phone. Doesn’t seem that there’s much room for improvement here by applying an EHR decision support system.  

The graph in this slide shows the % adherence to best practice for the prescription of nephrotoxic drugs, for all prescribers during the first three years of use of the EHR decision support system. The trend is essentially flat with about 40% adherence to best practice as defined by the EHR system. In that system, the gold standard for best practice was derived from the medical literature and then validated for every individual drug contained in the system. Considering how easy it was to confirm the best practice order presented by the EHR, and how much work is involved in order write an incorrect order, the users must have put in quite a bit of effort to get it wrong. Why would they do this? The result is not at all what I would have expected, taking into account their first class performance when it came to prescribing ACE inhibitors in the setting of heart failure.  

A few slides back I made reference to the fact that physician practice varies widely in terms of the numbers of verbal orders given per doctor. Over the 3 year period of this review, nearly 730,000 orders of all kinds were generated by clinical staff, about 175,000 of which were orders for medications. Of the 175,000 medication prescriptions, on average, 46% were actually written by the doctor himself or herself. Most of the orders actually written by the doctor were in ordersets, where one or two keystrokes result in an entire set of prescriptions being generated. On average, about 70% of all prescriptions written by the physician himself or herself were generated through these ordersets, which are built to each individual physicians’ personal specifications.

This slide and the slides which follow group the data so that we can examine the performance of those doctors who write more than 50% of their prescriptions themselves, versus those doctors who write less than 50% of their prescriptions themselves. All prescriptions not actually written by the prescribing doctor are given to the nursing staff as a verbal order, who then transcribe the orders into the decision support module, generating a communication with the pharmacy and an update of the Medication Administration Record entry.

This slide examines nursing performance in regards to best practice in the setting of renal failure. Each lavender bar indicates % adherence to best practice for each of the three years covered in the study. The dark lavender bars represent the performance of nurses transcribing orders for patients of doctors who write more than 50% of their own orders, themselves. The light lavender bars represent the performance of nurses transcribing orders for patients of doctors who write less than 50% of their own orders themselves. Any individual nurse could be represented in either or both bars for the same year, as we are examining their behavior relative to the doctor responsible for the patient they are caring for.

In the previous slide we saw that across the board adherence to best practice was flat at about 40% for all prescribers across three years. In this graph we see that the adherence to best practice by nurses was worse than the average, just reaching 15% in the third year. Remarkably, therefore, the nurses went to the trouble of changing 85% of orders recommended by the EHR system to the wrong orders. Of interest, but not statistically significant, in each year of the study, we see that nurses transcribing verbal orders for the group of doctors who write more than 50% of their own orders, performed better than when they transcribed orders for doctors who used the application themselves as little as possible.

This slide was constructed similarly to the slide we most recently reviewed, except now we are looking at performance of the doctors themselves. The light blue bars represent best practice performance in the setting of renal failure by the group of physicians who write less than 50% of their own orders themselves. The dark blue bars represent best practice performance by the group of physicians who write more than 50% of their own orders themselves.

Performance by the doctors writing less than 50% of their own orders is flat, at about 40% adherence to best practice over 3 years. Though better than the nurses performance when given verbal orders by these same docs, the effort required to have written the wrong order 60% of the time must have been considerable, taking into account the extra time it would have taken at the keyboard to accomplish that.

Performance by the doctors writing more than 50% of their own orders is remarkable. Not only did these docs start off with a better record than that of their colleagues when the EHR system was first implemented, but as time progressed they followed the best practice recommendations to a greater and greater extent, hitting 85% compliance during the third year. A very respectable record.

In this slide we see the same best practice data for the high-performance doctors compared to all of the other groups.

So what can we conclude?

In the setting of renal failure, doctors who write their own prescriptions for their patients are better for their patients than doctors who have the habit of delegating order entry responsibilities to nurses.

Doctors who depend on nurses to transcribe more than 50% of their prescriptions are responsible for twice as many incorrect prescriptions for nephrotoxic drugs as their physician colleagues who write more than 50% of their patient prescriptions themselves.

Nurses who transcribe prescriptions for nephrotoxic drugs avoid best-practice solutions 80 to 90% of the time. In fact, nurses who transcribe prescriptions for nephrotoxic drugs perform worse with respect to best practice in renal failure than either prescribing-class of physician. 

What does an EHR have to do with any of this? From one perspective, we might conclude, “not much,” but we would be wrong. It’s true that overall adherence to best performance in the setting of renal failure is flat over three years, but at least it’s flat at 40%. That’s a lot better than 0, and in the absence of an EHR-based decision support system it would essentially be 0. Well, maybe not 0, as the pharmacist would try to validate all orders in a timely fashion, but in the absence of an EHR, would have to access one or another informatics system manually to find relevant test results one at a time, or find them in a paper chart, or would have to go to the care unit to get patient’s height and weight which in the absence of an electronic system either are not performed or available only by locating the paper chart, itself perhaps accompanying the patient on a two hour trip to radiology for an imaging examination.

What else does the EHR bring to the table? It allows us to generate a quality control audit based on the real practices of individual staff, in the course of a single day, essentially a real-time response to a query directed by hospital quality-assurance staff. In the case of renal failure cases treated at our hospital, it shows us that the relationship between nursing and physician staff is probably of the one-way variety, and that an important effort needs to be made so that nurses and doctors feel a part of the same care delivery team. It is just not best practice for one group to simply delegate to the other. We see perhaps the hint of movement in the right direction in the case of the nurses transcribing verbal orders for the group of physicians who write most of their orders themselves. One explanation for nurses slightly improved adherence to best practice when transcribing orders for that particular group of docs is that they actually reported the decision support recommendation to some of these docs, who then supported the EHR recommendation. 

If prescribers wrote for medications with no contraindications, no side effects, and no dosing limitations, then anybody could write an order. But medications don’t work that way, and the literature shows despite the incredible advances in medical knowledge over the past two or three decades, despite the academic credentials of professional staff, simple oversights occur time after time after time in the real world. These are the oversights that keep JCAHO in business, the oversights that resulted in the development of the Leapfrog Group Report, and the oversights that are at the origins of a substantial fraction of medico-legal risk.

I’d like to stop here, as many of you have as much or more experience than I do with the implementation and use of EHR. I’ve presented a slice of data which could not possibly have been collated by hand from a paper record, nor for that matter, from any other system other than one which supports comprehensive clinical care documentation at every level of the clinical care process.

If there are any questions, I’d be pleased to address them. Thank you.