Clinical Decision Support Systems (CDSS)
Traditionally, CDSS meant computerized drug alerts and reminders to perform preventive tests as part of computerized physician order entry (CPOE) applications. Most of the studies in the literature evaluated those two functions. However, according to Hunt, CDSS is “any software designed to directly aid in clinical decision making in which characteristics of individual patients are matched to a computerized knowledge base for the purpose of generating patient specific assessments or recommendations that are then presented to clinicians for consideration.”53 Therefore, CDSS should have a broader definition than just alerts and reminders. Two 2005 papers addressed the effects of CDSS on clinical care. Garg and co-authors concluded that overall, CDSS improved performance in 64% of the 97 studies but only 13% of the 52 studies analyzed reported improvement in actual patient outcomes.54 Kawamoto et al. looked at those factors that contributed to the success of CDSS: automatic CDSS that was part of clinician work flow; recommendations and not just assessments; provision of CDSS at the point of care and computer-based CDSS (not paper-based). When these four features were present, CDSS improved clinical care about 94% of the time.55 According to a 2009 article, clinical decision support by nine commercial EHRs was extremely variable and tended not to offer choices.56 Clearly, the most sophisticated CDSS are developed at medical centers with home grown EHRs and a long record of extensive HIT adoption. With Meaningful Use criteria, certified EHRs will have to conform to CDSS standards which may reduce variability. Sheridan and Thompson have discussed various levels of CDSS: (level 1) all decisions by humans, (level 2) computer offers many alternatives, (level 3) computer restricts alternatives, (level 4) computer offers only one alternative, (level 5) computer executes the alternative if the human approves, (level 6) human has a time line before computer executes, (level 7) computer executes automatically, then notifies human, (level 8) computer informs human only if requested, (level 9) computer informs human but is up to computer and (level 10) computer makes all decisions.57 Most EHR systems may offer alternatives and provide reminders but make no decisions on their own. With artificial intelligence and natural language processing becoming more sophisticated, this could change in the future. Table 4.2 outlines some of the clinical decision support available today. Calculators, knowledge bases and differential diagnoses programs are primarily standalone programs but they are slowly being integrated into EHR systems.
Numerous digital medical resources are being integrated with EHRs. As an example, the American College of Physician’s PIER resource is integrated into Allscript’s Touch Chart.58 The comprehensive online reference UpToDate has been integrated into six EHRs and has an option to connect to other EHRs via an API.59 iConsult (offered by Elsevier) is a primary care information database available for integration into EHRs. Diagnostic (ICD-9) codes can be hyperlinked to further information or users can use infobuttons. Other products such as Dynamed, discussed in the chapter on online medical resources are available as infobuttons. Figure 4.3 shows an example of iConsult integrated with the Epic EHR.60 Another interesting integrated knowledge program is the Theradoc Antibiotic Assistant. The program integrates with an inpatient EHR’s lab, pharmacy and radiology sections to make suggestions as to the antibiotic of choice with multiple alerts. Clinicians can be alerted via cell phones, pagers or e-mail. Other modules include Adverse Drug Event (ADE) Assistant, Infection Control Assistant and Clinical Alerts Assistant.61 A study in the New England Journal of Medicine (NEJM) using this product showed considerable improvement in the prescription of appropriate antibiotics resulting in cost saving, reduced length of stay and fewer adverse drug events.62
Table 4.2 Clinical decision support
|Type of CDSS||Examples|
|Calculators||Medcalc 3000®, eCalcs|
|Trending/Patient tracking||Flow sheets, graphs|
|Medications||CPOE and drug alerts|
|Order sets/protocols||CPGs and order sets|
|Radiology CDSS||What imaging studies to order?|
|Laboratory CDSS||What lab tests to order|
|Public health alerts||Infection disease alerts|
It is likely with time calculators will be embedded into all EHRs, particularly in the medication and lab ordering sections. Figure 4.4 shows a calculator program that integrates more than 30 common calculations into a commercial EHR (Allscripts). The fields are automatically calculated and results can be added to the encounter note.63 Note that the figure shows a Framingham cardiovascular risk score determination. Important calculations, such as kidney function (creatinine clearance) should be calculated and available on all patients, particularly when prescribing drugs that are excreted by the kidneys or imaging contrast agents that can be toxic to the kidneys.
Flow sheets, graphs, patient lists and registries.
The ability to track and trend lab results and vital signs, for example, in diabetic patients will greatly assist in their care. Furthermore, the ability to use a patient list to contact every patient taking a recalled drug will improve patient safety. Registries will be covered in more detail in the disease management chapter.
Medication ordering support.
Decision support as part of CPOE possesses several rules engines to detect known allergies, drug-drug interactions, drug-condition and drug-food allergies, as well as excessive dosages. As EHRs and CPOE mature, they will factor in age, gender, weight, kidney (renal) and liver (hepatic) function of the patient, known contraindications based on known diagnoses, as well as the pregnancy and lactation status. Incorporation of these more robust features is complicated and best implemented at medical centers with an established track record of CDSS and CPOE development. As has been pointed out, there are programs that improve antibiotic ordering based on data residing in the EHR.64 Computerized drug alerts have obvious potential in decreasing medication errors but have not been universally successful to date. According to a systematic review by Kawamoto et al., successful alerts need to be automatic, integrated with CPOE, require a physician response and make a recommendation.55 Four studies have been published from the Brigham and Women’s Hospital showing mediocre compliance, even for black-box type warnings.65-68 An excellent review by Kuperman et al. describes basic and advanced medication-related CDSS.69 Further information about alerts is included in the chapter on patient safety.
Computerized reminders that are part of the EHR assist in tracking the yearly preventive health screening measures, such as mammograms. Shea performed a meta-analysis and concluded that there was clear benefit for vaccinations, breast cancer and colorectal screening, but not cervical cancer screening.70 A well-designed system should allow for some customization of the reminders as national recommendations change. Reminders are not always heeded by busy clinicians who may choose to ignore them. As a possible solution, preventive reminders could be reviewed by the office nurse and overdue tests ordered prior to the visit with the physician.
Order sets and protocols.
Order sets are groups of pre-established inpatient orders that are related to a symptom or diagnosis. For instance, users can create an order set for pneumonia that might include the antibiotic of choice, oxygen, repeat chest x-ray, etc. that saves keystrokes and time. Order sets can also reflect best practices (clinical practice guidelines), thus offering better and less expensive care. Over one hundred clinical practice guidelines are incorporated into the electronic health record at Vanderbilt Medical Center.71 For more information on order sets readers are referred to this reference.72
Dxplain is a differential diagnosis program developed at Massachusetts General Hospital. When clinicians input the patient’s symptoms it generates a differential diagnosis (the diagnostic possibilities). The program has been in development since 1984 and is currently web-based. A licensing fee is required to use this program. At this time it cannot be integrated into an EHR.73 In spite of the potential benefit, an extensive 2005 review of CDSSs revealed that only 40% of the 10 diagnostic systems studied showed benefit, in terms of improved clinician performance.74 Artificial intelligence continues to improve so it is likely that EHRs will have the ability to assist with differential diagnosis in the future.
Physicians, particularly those in training, may order imaging studies that are either incorrect or unnecessary. For that reason, several institutions have implemented clinical decision support to try to improve ordering. Appropriateness criteria have been established by the American College of Radiologists. Massachusetts General Hospital has had radiology order entry since 2001 and studied the addition of decision support. They noted a decline in low utility exams from 6% down to 2% as a result of decision support.75
It should be no surprise that clinicians occasionally order inappropriate lab tests, for a variety of reasons. It would be helpful if clinical decision support would alert them to the indications for a test, as well as the price. A Dutch study of primary care demonstrated that 20% fewer lab tests were ordered when clinicians were alerted to lab clinical guidelines.76
Public Health Alerts.
The New York Department of Health and Mental Hygiene used Epic EHR’s “Best Practice Advisory” to alert New York physicians about several infectious disease issues. The EHR-based alert also hyperlinked to disease specific order sets for educational tips, lab and medication orders.77 How well clinicians use CDSS programs such as those discussed, remains to be seen. They will have to be intelligently designed and rigorously tested in order to be accepted. For more information on CDSS, readers are referred to the resources cited in these references.78-82
Approximately five billion prescriptions are written annually in the United States and until about 2009 the majority were still paper-based.83 This trend has changed dramatically, due to increased EHR adoption; such that by the end of 2012, 87% of electronic prescribing was EHR based, 69% of office-based prescriptions were electronic and 93% of community pharmacies were connected to the Surescripts network.84 The potential multiple advantages of e-prescribing are as follows:
- Legible and complete prescriptions that help eliminate handwriting errors and decrease pharmacy “callbacks” and rejected scripts
- Abbreviations and unclear decimal points are avoided
- The wait to pick up prescriptions potentially is reduced
- Fewer duplicated prescriptions
- Better compliance with fewer drugs not filled or picked up
- Potential to reduce workload for pharmacists
- Timely notification of drug alerts and updates
- Better use of generic or preferred drugs
- The ability to check plan-level and patient-level formulary status and patient copays
- E-prescribing can interface with practice and drug management software.
- The process is secure and HIPAA compliant.
- It is the HIT platform for future clinical decision support, alerts and reminders. It could integrate decision support related to both disease states and medications.
- Digital records improve data analysis of prescribing habits.
- Programs offer the ability to look up drug history, drug-drug interactions, allergies and compliance.
- While entering an e-script is slower than writing a paper script, clinicians have options to speed up the process like batch refills and choosing from lists of drugs most commonly prescribed in a practice.
- Provides a single view of prescriptions from multiple clinicians.
- Applications have the ability to check eligibility, co-pays and it can file drug insurance claims.
- Overall, e-prescribing is associated with reduced cost of prescribing.85
It is not thought that simply switching from paper to electronic prescriptions will improve patient safety; it will require clinical decision support systems (CDSS) that alert and educate potential medication issues. Perhaps the most important CDSS is the reminder that a patient has a confirmed allergy to a drug, thus preventing a potential serious reaction. It is most helpful if the actual details of the allergy are listed (e.g. Sulfa family, anaphylaxis 2012). The next important CDSS feature is drug-drug interaction determination. In elderly patients on multiple medications it is particularly important to understand the effect of one drug on another. Notification of an interaction will usually cause the prescribing physician to reduce the dose of one drug or make another safer choice. There are many other types of CDSSs that might be important associated with e-prescribing. Drug-condition/disease alerts might remind a physician that drug A is not safe in a pregnant woman. Reminders about dosages out of range (too high or too low), age or BMI extremes would be very valuable, particularly with toxic drugs such as chemotherapy medications. Reminders about duplicate drugs and drugs prescribed by other physicians are also very important. As electronic health records become smarter by using rules engines and artificial intelligence users can expect alerts about potential prescribing problems based on liver or kidney problems and other considerations. Eventually, there may be summary alerts based on age, gender, BMI, liver/kidney function, etc., such as “This patient is at risk of drug side effects, recommend Lisinopril dose reduction by 50%.” Another example would a reminder about medications with sedating properties in the elderly. As noted previously, the vast majority of e-prescribing now takes place as part of the electronic health record. There is evidence that e-prescribing as part of an EHR reduces medication errors but many questions remain.86 Some of the issues with CPOE in this chapter and the chapter on patient safety have been addressed. They following are some of the issues or challenges associated with e-prescribing:
- Alerts, in general, are viewed as nuisances by physicians, unless they are very specific, highly important and are educational.87
- One study evaluated the pharmacist’s perspective and disclosed unique new e-prescribing issues: incorrect drugs, doses and patient instructions continue to occur; in spite of an electronic process prescribing delays persisted. They recommended that only clinicians forward e-prescriptions, clinical decision support should be used, scripts should be sent together (bundled); software standardization would be helpful and there should be a mechanism to message physicians about issues.88
- A study of 3850 outpatient electronic prescriptions reported in 2011 revealed an error rate of 11.7%, with about a third having the potential to cause adverse drug events (ADEs). Two thirds of the prescribing errors were due to omissions of drug dose, instructions, etc. Actual ADEs were not reported.89
- A qualitative study of e-prescribing was reported in 2011 and recorded some of the existing issues physicians and pharmacists are facing:90
- The refill process had more problems and errors than the initial new prescription process and resulted in workarounds for both physicians and pharmacies.
- Some pharmacies don’t accept electronic scripts because they don’t want to pay Surescripts fees.
- Mail order pharmacies still lack consistent e-prescribing capabilities. Most of their refills are still done by fax.
- Physicians write sigs (instructions) that aren’t patient friendly and pharmacists have to rewrite them.
- Physicians often receive duplicate requests from pharmacies for a variety of reasons.
Over 112,000 health care professionals use Practice Fusion for their EHR
See for yourself why our EHR was ranked #1 for ease of use.