Delivering standard of care (SOC) medicine is becoming increasingly more complex. The knowledge explosions in clinical medicine and molecular biology have made mastery of all but the most specialized fields of medicine beyond the scope of human ability. As this knowledge is applied to a given patient, by any number of caregivers, that patient’s medical history then develops similar challenges in scope, organization, analysis, and retrieval. Fitting a complex knowledge base to a complex patient would be difficult for a single caregiver. By using a team approach to care, with multiple specialists and multiple handoffs, the potential for deviations from SOC medicine increases yet again. Faced with the obligation to deliver the right care to the right patient at the right time, every time, contemporary medicine has turned to health information technology (HIT), and specifically the electronic medical record (EMR) to meet these challenges. A well-designed EMR has the potential to help organize and analyze patient data, link to appropriate clinical guidelines, and provide guidance to all members of the care team. The rules and algorithms that make this happen constitute the clinical decision support (DS) system (CDSS). A good CDSS should make it easier for a provider to make clinical decisions that will improve quality and enhance patient safety.

To formalize the many complexities of contemporary medicine, protocols have been developed. In addition to clinical guidelines, payors, administrators, credentialing bodies, governments, and patient advocate groups all have their expectations about the content and delivery of medical care. These protocols include standards for evaluation, diagnosis, and treatment, as well as ancillary functions such as patient education, billing, and other aspects of health care workflow. Quality, safety, and error are concepts that are closely related to protocols, but operationally, especially in the world of HIT, need to be distinct. Errors are deviations from the SOC, and represent unsafe situations with the potential for patient harm. In addition, errors leave the practitioner vulnerable to all types of regulatory, financial, and legal liability, the last of which is all too familiar to most obstetricians. “Quality” is the content of the protocol; eg, all term group-B streptococcus (GBS)-positive parturients will receive antibiotic prophylaxis. “Safety” is the details of executing that protocol in a specific care environment; eg, who sends the GBS results to labor and delivery, who receives the results, who charts the results, who checks the results, who orders the antibiotics, and who administers the antibiotics. Quality is global; safety is local. Quality is static and after the fact; safety is dynamic and in real time. An error in quality (eg, this GBS-positive patient did not get antibiotics) gives no details about why, and little opportunity for correction, whereas identifying an error in safety (not received, not charted, not recognized, not ordered, not given) makes it easy to see who needs to do what to get back on protocol. DS is any type of tool that helps a caregiver to stay on protocol, by sending a message when a deviation from protocol has occurred. Therefore, CDSS is integral to patient safety as it is focused on changing provider behavior at the point of care, thereby ultimately improving quality.

Multiple paper-based methods have been used to help providers follow a protocol. Preprinted templates, checklists, and order sets are all examples of ways to make it easier to evaluate, diagnose, and treat patients in a standardized way. Quality is what is included in the template or checklist; safety is how often it is followed. With the advent of EMR, tools for protocol adherence can be much more sophisticated. While the basic tasks (elements) of patient care remain the same, the functionality and design of these elements can be thoughtfully integrated into both the substance of the EMR, and into patient care workflow in ways that make it easy and obvious for caregivers to increase their compliance with established standards. In fact, given the exploding complexity of medical care, HIT supporters maintain that a CDSS may be the only way to provide this assurance of patient safety. This support will hopefully be able to help close the large and well-documented gap between the best evidence and the actual practice of medicine.

To understand a CDSS, one must first consider the fundamental building blocks of any EMR. An EMR has several main purposes, including the ability to document, communicate, analyze, monitor, and report on a provider’s entries into the medical record. What transforms the EMR from a paper- or text-based record is the use of structured, coded data fields and a controlled medical vocabulary. In the paper world, obstetricians use many different words and phrases to indicate that a fetus is not tolerating the stress of labor, and that a cesarean section is warranted. In the electronic world, this statement must be confined to a controlled medical vocabulary–a limited and specific number of phrases with agreed-upon definitions. In addition, the statement must be located in a specific field or location in the EMR–a structured, coded field. Without these 2 elements, an EMR is just a series of text files, and, with the exception of the nascent field of natural language processing, does not lend itself to further analysis. Once structured coded fields are in place for documentation, the computer can manipulate those fields for other functions required for patient care. From the aspect of patient safety, these data elements can be used to invoke rules or protocols that define the SOC for a given situation. If there is a deviation from the SOC, additional actions can be suggested to the care team so that the SOC can be met.

One of the most basic tasks of a CDSS is the ability to determine voids in the database. A rule to encourage antibiotic coverage of GBS-positive patients will never fire if the GBS data field is blank. Recent statistics from the Joint Commission suggest that inadequate patient assessment is a factor in >50% of adverse sentinel events. Having an EMR is no protection against poor-quality data, and specifically the EMR may not always contain sufficient amounts of coded data to drive DS. Therefore, a good CDSS will alert the provider to data voids. These requests can be general (all patients need a weight and a blood pressure recorded on initial assessment), or driven by specific time intervals (all patients on magnesium need a progress note every 2 hours) or conditions (all laboring patients need a GBS result).

After the fundamental task of ensuring an adequate assessment of the patient, CDSS can add additional layers of support. Many decisions in clinical care are driven by an understanding of exactly what the patient’s problems are, but establishing an accurate problem list remains a challenge in most inpatient workflows. In the Harvard Medical Practice Study, diagnostic failures represented the fifth most common type of adverse event, accounting for 8.1% of all events. A CDSS can compare entered signs and symptoms against defined disease profiles, and make suggestions for possible diagnoses. Full-blown diagnostic programs have not been successful in clinical use, but a rule to suggest the problem of preeclampsia in a parturient with new-onset hypertension and proteinuria is not a difficult task.

Using a DS algorithm to interpret a fetal monitoring record has been a subject of investigation for decades. Several attempts have demonstrated equivalence of such algorithms to human interpretation, however, independent demonstration of clinical benefit has not been shown. This remains an area of active investigation. The simpler alarm bells that monitor events such as bradycardias have likewise not demonstrated clinical benefit, although they may aid in nursing workflow.

The problem list can then be used to drive other rules and protocols. These include suggestions for tests and therapies, as well as strategies for monitoring the condition of the patient over time. Most high-acuity units have protocols about the frequency and content of patient assessment. An unaugmented patient with a Eunice Kennedy Shriver National Institute of Child Health and Human Development category-1 tracing needs much less frequent reevaluation than a patient on oxytocin with a category-2 tracing. CDSS rules can send reminders if progress notes are not entered in a timely fashion. Outpatient EMRs can also benefit from CDSS. In addition to the functions mentioned above, reminders for preventive health care activities and follow-up of ancillary testing and referrals can be monitored, with reminders sent to both the primary care physician and the patient. By using findings and problems to guide actions and reassessments, a CDSS can help providers meet the SOC for even the most complex situations.

Another important aspect of CDSS is error recovery. Most errors in clinical practice that lead to bad outcomes are not single events. Instead, an abnormal finding is ignored, a confirmatory finding is not assessed, a test is not ordered, a reassessment is not done, a diagnosis is not made, and over time, these errors compound into a serious adverse event, a model well described by Reason. DS systems need to be able to monitor this deviation from protocol over time. By increasing their level of alerting, and by moving up a chain of command to notify other care team members, a high-level CDSS has the potential to significantly decrease adverse events.

Communication, as noted above, is a fundamental purpose of an EMR, and an increasingly error-prone task. Care teams and frequent patient handoffs, as well as increasing levels of specialization, make it difficult for a given caregiver to have all the data necessary to provide optimum care. As the volume of documentation increases, sorting through accumulated results can be more challenging in the electronic world than in the paper one. Using the principles of visual design mentioned below, reports can be designed that populate with relevant data for a user’s role and task, saving time and ensuring completeness.

Having a broad array of targets for CDSS intervention expands the scope of patient safety initiatives. However, if every protocol deviation resulted in an intervention with a hard stop, the system would be unusable. Consider the example of a progress note for a patient on magnesium. Not letting a note be completed until the provider searches for, locates, and fills in the field for deep tendon reflexes would be a recipe for disaster. Including that field in a template for the note makes the CDSS an ally instead of an antagonist. The real challenge of designing a CDSS intervention is to have the intervention be as unobtrusive as possible. Interventions that offer advice, are available in real time at the point of care, and do not interrupt workflow are most likely to be accepted by physicians. Systems that have multiple levels of presenting an intervention can be adapted to institutional customs, resources, and workflows. The presentation of templates and order sets lets the provider decide what to choose, whereas required fields, alerts, and hard stops are examples of interactions that interrupt the provider. Nonetheless, sometimes an intrusive alert will be called for. Alerts that give choices (“order anyway/cancel order”) are preferable to ones that are always dismissed in the same way (“OK”). Not only is user acceptance greater, but the cumulative data give valuable information to HIT management about provider behavior. An additional enhancement for a hard stop gives the provider the ability to move forward by providing a reason why, in this particular clinical case, a protocol deviation has been chosen. Reporting and analysis of these reasons can aid in CDSS design, as well as identify opportunities for provider education.

Having the ability to alter the way a CDSS intervention is presented at the design level is wonderful. Being able to alter the presentation on the fly, depending on clinical context, provides a much higher level of support. For example, an estimation of fetal weight (EFW) is a recommended element of the physical examination of a laboring patient. In a vertex, 34-week patient, the need for this data point is debatable. In a 40-week patient, the EFW is more important. In a 41-week patient who has gestational diabetes and has been at 5-cm dilation for the last 6 hours, EFW is an element of extreme importance. A CDSS that keeps EFW in the background for the 34-week patient, but requires immediate attention to that data field in the 41-week patient would fulfill this type of functionality. The more important a given data element is (or becomes), the more emphatically it should be presented to the caregiver. A system that can progress from having a data field available, to making it prominently available, to alerting about its absence, to making it required, all on the basis of clinical context, would characterize a very high-level CDSS.

The visual design of information deserves as much planning, analysis, and reiteration as the more intrusive expressions of CDSS function. In fact, attention to visual design may obviate the need for alerts and pop-ups that interrupt workflow, tend to annoy users, and can produce alert fatigue. At a fundamental level, the use of colors, different font sizes, highlighting, and screen placement may serve to call attention to critical information. At a higher level, anticipating informational needs by careful observation of workflow, and automatically presenting relevant fields for data review can keep the CDSS in the background, and increase the ability of providers to work without interruptions. Doctors count clicks, and the more onerous data entry or retrieval becomes, the higher the chance for error. Visual summaries of row and column information–graphs, call-outs, drill-down capability–are a specific example of the so-called widget, a small window on the screen that can aggregate content from the database for a specific purpose, and collectively can form a “dashboard,” giving a dynamic view of a patient’s condition. The fact that an EMR contains ALL of a patient’s medical data is seldom a positive attribute; most providers only want to know about the data relevant to the task at hand. Well-designed graphic summaries can make this task easy.

The interactions from a CDSS may also be modified by the role of the user. Doctors may log on to the patient’s chart infrequently, whereas nurses may log into a given patient’s chart several times an hour. Messages from the CDSS to the doctor may need to be sent to the nurse first, with different modes for interaction and context. In addition, the rationale for DS rules needs to be assessable by the user. Links to primary literature and hospital protocols and procedures, as well as the ability to search the literature in general, can offer crucial advice to clinicians when a clinical case does not seem to fit neatly into a typical pattern. Glossaries, instructions for tests, detailed drug information, and similar types of supporting information can help providers make better decisions.

The last key part of CDSS functionality is the ability to report on how the system actually performs. How often do users fill in essential data fields? How well are problem lists maintained? How often do alerts fire, and do they change provider behavior? Understanding the answers to these questions allows the HIT staff to adjust the CDSS to meet the actual needs of the providers.

In summary, the functional aspects of a CDSS are what integrate the protocols with the actual workflow of the practitioners. Making this integration as unobtrusive as possible is a key factor in the user acceptance of CDSS. Having a robust tool set to adjust the methods of CDSS intervention is a key part of the implementation process; having this adjustment made according to clinical context is even better. In addition, a significant reporting function as to how the CDSS is actually operating, as well as the results of provider actions in response to the CDSS suggestions, can guide rational reiteration of the CDSS.

Computerized provider order entry (CPOE) has been heralded as one of the mainstays of using HIT to improve patient safety. Medication errors remain a significant source of error and patient harm. However, the literature supporting the benefit of CPOE remains mixed. Multiple systematic reviews have shown inconsistent results for preventing adverse drug events. The academic literature is hampered by imperfect methodology, and further constrained by mostly focusing on homegrown systems; commercially available products have had essentially no prospective randomized validations of their safety or effectiveness. Perhaps most revealing is a 2010 study by Metzger et al that evaluated the safety performance of CPOE at 62 hospitals. Test patients were entered into the hospitals’ production systems, and then a series of orders were placed on those patients. Cumulatively, the CDSS detected only 53% of orders that would have caused a fatality, and 10-82% of orders that would have caused a serious adverse drug event. Drug-allergy and drug-drug interactions were detected with a higher rate, but contraindications involving the patient’s age, diagnoses, and laboratory values were picked up much less frequently. Marked variation in effectiveness was present even among users of the same commercial system, pointing out the importance of customization of rule sets.

Using CDSS to make sure that a drug is safe to give should be fairly straightforward. Checking to see if a drug is actually indicated brings a new set of challenges. Currently, a drug’s indications are not mapped to a standardized vocabulary, so that a CDSS module to compare a prescribed drug’s indications with the patient’s problem list is difficult to construct. Drug companies may soon be required to codify a drug’s indications with a controlled medical vocabulary that could then be easily compared to a patient’s problem list for drug-disease CDSS. An additional and much more critical challenge is the ability for a CDSS to suggest a drug’s discontinuation as the patient’s clinical situation changes over time. Stopping heparin for heparin-induced thrombocytopenia or stopping oxytocin when uterine tachysystole occurs would be examples that would have critical impacts on patient safety.

For a CDSS intervention to change the care of a given patient, it must be given in real time. However, many sources of error in health care are thought to be system issues that have the potential to endanger many patients over time. To improve the entire system, aggregated data may provide insights. Without a significant reporting capability, EMRs become black holes of data input, and the provider becomes little more than a data entry clerk for the billing department. A robust CDSS should have the capability to report cumulative statistics to both the provider and the management team. Making the CDSS fit the protocol is easy; making it fit the culture and workflow of any particular institution is a work in progress. Understanding which rules fire under which circumstances provides the data for rational revision of the rules. Merging cumulative CDSS data with patient outcomes can focus attention on areas of unmet need, and drive revision of the CDSS itself. Reporting of significant protocol deviations to a risk management department may alert that team to possible adverse events, or cases for root cause analysis. Automatic identification of adverse drug events identifies many more events than self-reporting or manual chart review. All of these reports, brought back to management, may suggest ways to provide system changes to make a work process safer. In addition to periodic reports, business intelligence tools have the ability to create dashboards that can monitor aggregate data in real time.

Despite the attempts to increase patient safety with EMR/CDSS, the introduction of these tools brings with them new sources of error. An early paper pointed out numerous potential errors in a CPOE system. With complex systems, workflows can easily become work-arounds, bypassing intrinsic safety features. Provider’s knowledge of the system represents another source of error. In a recent survey, practitioners overestimated the amount and type of CDSS present in the system they were using. Suggested diagnoses or interventions might be accepted by the practitioner, starting a cascade of errors based on blindly accepting that the computer must be right. Potential for these types of errors points out the need for constant monitoring of events and outcomes, revision of the system, and education of the users.

Every patient care environment has its own history, culture, resources, and special needs. A CDSS is the glue that binds a unique patient care environment to a standard protocol, so the protocol can be consistently executed, and patient safety achieved. To fit an SOC to a unique care delivery system will require significant customization. Individualized systems will require more resources to build and maintain, and will present an additional challenge to those providers who see patients in multiple environments. However, there is little evidence to support the use of an out-of-the-box CDSS to improve patient safety. Much of the published research on the benefits of EMR systems comes from institutions where the EMR was developed. The feelings of engagement and ownership of a locally developed solution may go a long way toward acceptance and success. In fact, when CDSS applications are applied to institutions where they have not been developed, the evidence of their effectiveness decreases considerably. Strong support from leadership, and a team that includes experts in clinical knowledge, HIT, and human-computer interaction, can get a system up and running. A tool set that includes the ability to alter the design and adjust the level of intervention must be available at least at a programmatic level, to be able to alter these parameters dynamically is a huge advance in sophistication. Close monitoring of system behavior and clinical outcomes can provide the basis for rational reiteration of both elements and functionality. By combining customization, a robust tool set for doing so, and the willingness and resources for rational reiteration, a CDSS can have the best chance for improving patient safety.