Isabel (www.isabelhealthcare.com) is an award-winning and validated diagnosis decision support system, conceived 7 years ago, as a direct response to the near-fatal misdiagnosis of Isabel Maude, who developed necrotizing fasciitis, a well described complication of chicken pox. Isabel was seen by both the family physician and the local hospital’s ED, all of whom failed to recognize the typical clinical features of necrotizing fasciitis, and sent the patient home.

Isabel’s life-threatening ordeal highlighted the need for an advanced diagnosis reminder and knowledge system-one that would be fast and easy to use in the clinical workflow.

The Isabel system was created in 2001 by physicians to offer clinical decision support at the point of care. It has been extensively validated and has been shown to enhance physician’s cognitive skills and thereby improve patient safety and the quality of patient care.

What is the Isabel Diagnosis Checklist System (IDCS) and Knowledge component?

What is the IDCS not intended for?

How does the Isabel system differ from other systems such as Dxplain and QMR?

Is the order of appearence of a diagnosis in Isabel the order of importance?

I have access to UpToDate and MDConsult, why do I need the Isabel system?

Is the Isabel system an informational tool or a clinical tool?

In what healthcare setting is Isabel intended for?

Is the Isabel system an educational tool?

Will the Isabel system encourage me to think less for myself?

I do not see many complicated medical cases, why would I use the Isabel system?

Will the Isabel system cause unnecessary tests to be ordered?

Who creates the Isabel content and how often is the system updated?

Not only do I have limited time to learn a new system, but I do not have enough time to use a system during practice.

I have a PDA. Can I use Isabel?

What validation have Isabel and the other diagnosis decision support systems undergone?

Isabel Validation - Independent Studies:

Isabel Validation - Simulated & Clinical Outcome Studies.

(i). Has Isabel 's system accuracy been tested?

(ii). Has the Isabel system's impact on decision making been examined?

(iii). What was the impact of Isabel on clinical decision making in an experimental setting?

(iv). What is the impact of Isabel on decision making in a clinical setting?

Has Isabel been interfaced with Electronic Medical records (EMR) ?

How does Isabel Clinical Alert and Monitoring system (ICAM) deliver real-time epidemiology ?

Can Isabel suggest drugs that might cause the clinical features in a patient ?

How does Isabel assist in Bioterrorism preparedness?

What is the magnitude of diagnosis error?

What are patient's perceptions of diagnosis error?

Do cognitive factors (premature closure) contribute to diagnosis error?

How can Isabel assist in minimizing cognitive factors contributing to diagnosis error?

Do physicians know when their diagnoses are correct? How likely are physicians to use diagnosis decision support systems?

The Isabel clinical decision support system has two main components: the IDCS and Knowledge component. For a given set of clinical features from a patient, the IDCS instantly provides the clinician with a checklist of likely diagnoses and causative drugs. The Isabel taxonomy contains over 11 000 diagnoses and over 4000 drugs. The Knowledge component goes one step further: it harnesses diverse medical knowledge related to each of the diagnostic suggestions and drugs. It also searches an extensive knowledge base to find specific answers to clinical questions arising in the workflow.

The IDCS is intended to be used as a near-patient decision support tool. It is designed to be used by clinicians to provide a checklist of likely diagnoses and drugs for a patient's clinical features. In this fashion, it aims to reduce diagnostic errors and improve patient care.

The Isabel system is not an electronic medical record. The IDCS is not intended to be used as an oracle to solve medical problems. Nor is it to be used as an 'expert system' that is considered superior to the clinicians involved in the patient's care.

Isabel differs from rules based DDSS like QMR, Iliad, Dxplain, DiagnosisPro, PKC in that it uses statistical natural language processing [SNLP] software and is thus able to handle unstructured data. Entry/extraction of data is therefore very quick. Isabel searches the natural language of established textbooks. A study of QMR and Iliad showed "Each case took from 20 to 40 minutes to input" - How well does decision support software perform in the emergency department? PKC handles one chief complaint at a time and is designed for data to be entered by the patient. A recent study - A randomized outpatient trial of a decision-support information technology tool of PKC showed that "average Coupler session took approximately 18 minutes of employee time to coordinate." In contrast, on average the data entry time for Isabel on a stand alone basis [as opposed to extraction from EMR] has been shown in studies to be less than a minute-basically the time its takes to type 3-4 clinical features.

Dxplain and Quick Medical Reference (QMR) were developed two decades ago as 'expert systems' that aimed to solve clinical conundrums. They used a complex network of clinical findings and disease names within their database. Users entered a patient's clinical features through a controlled vocabulary. Diagnostic results were ranked by probability. In contrast, Isabel is not an 'expert system'; it simply aims to remind clinicians of key diagnostic possibilities. Results are arranged by body system. The database is merely a collection of textual disease descriptions. The user interface accepts natural language entries. Decision making is still left to the physician. Rules based systems like D-xplain give the user a ranked list of diagnoses. It’s understandable that clinicians balk, when presented with a system that seems to devalue their clinical education and experience.

Ease of data entry: Isabel has the ability to handle natural language / free text. Entry of data is therefore very quick [the time it takes to type clinical features into Isabel’s query box] and Isabel searches the natural language of established textbooks. Rules based systemsmanually link each clinical feature to a disease and this link/vector is given an evoking strength [ES] e.g.:

• Cough is linked to bronchitis with an ES of say, 5
• Cough with expectoration is linked to bronchiectasis with an ES of say, 7
• Cough with blood-tinged sputum is linked to Good pastures disease with an ES of say, 6
• Cough with frank blood (hemoptysis) is linked to Pulmonary TB disease with an ES of say, 8

Therefore with rules-based systems you have to answer a series of questions for each clinical feature. As a result, a study of QMR and Iliad showed "Each case took from 20 to 40 minutes to input" this gives an idea as to the time taken to obtain a differential diagnosis - DiagnosisPro & Dxplain.

Rules-based DDSS were designed and promoted as "expert" systems rather than "reminder" systems - the other reason why these are not as ubiquitously used as the enthusiasts would have liked. Dxplain gives the user a ranked list of diagnoses. Isabel is designed to enhance the knowledge and cognitive skills of the physician ["learned intermediary" - Jeremy Wyatt] at the bedside - to provide a checklist of reasonable, relevant diagnoses for the physician to consider.

The order of appearance of a diagnosis is not the order importance - physicians have to use their clinical judgment to decide - "which of these diagnoses am I going to investigate and crucially which am I going to treat". To rank order or not has been a debate since we first launched Isabel in 2002. We have taken advice from several health informaticians, patient safety experts and senior physicians and continued not to rank order them as we believe that this is asking too much of the system. There will always be information on the patient which has not been entered and therefore it would not be credible for the system to do this. Our view is that the clinician can quickly scan a list and decide for himself the probability. Graber ['Diagnostic error in Internal medicine'. Mark Graber MD, Department of Veterans Affairs Medical Center, Northport, NY. Arch Intern Med. 2005;165:1493-1499] found that lack of knowledge was not an important factor in diagnostic error and this why Isabel is called a reminder system-its is designed to remind the physician of diagnoses he knows about but may not of thought of; a safety net or check list. Isabel is not designed to be a system which tells the physician what to do-these have been tried and failed-but instead one that enhances the physicians skills.

UpToDate and MDConsult are large repositories of medical knowledge (journals, textbooks, reviews). It is possible to start a search within these systems only when a diagnosis label is available. Isabel starts at an earlier point in the clinical journey, and helps the clinician reach a diagnosis first, and then helps to mobilize medical knowledge related to the disease. However, if your institution has a subscription to these medical knowledge resources,Isabel can, in principle, link directly to them following diagnostic advice.

The IDCS is designed to be used in a clinical setting at the point of care. The Knowledge component can be used either in a clinical setting to quickly read up a Online Textbook, or in a library setting for further reading and ready reference.

It is anticipated that the Isabel system will be most useful in clinical settings with rapid turnover, maximum potential for diagnosis errors, and where diagnostic uncertainty is common. By this definition, most areas caring for acutely ill patients would be ideal settings for Isabel usage. In addition, it can be used in ambulatory settings and outpatient departments as a reference tool.

While the Isabel system is primarily a clinical decision support system, the educational aspect of its advice cannot be underestimated. Even if the diagnostic advice did not change practice for one patient, the knowledge gained from its advice might change the next patient's clinical journey. In addition, it has an invaluable role to play in medical education at any level to illustrate the art of diagnostic decision making.

Although the Isabel system will provide a set of diagnostic reminders for each query, it does not tell the clinician what to do, or even which diagnosis to further workup. Isabel leaves the responsibility of decision making to the user, but helps them by providing reminders as well as mobilizing relevant medical knowledge.

Isabel is not an 'expert system' designed to be used only in diagnostic dilemmas. As humans making complex decisions under stressful circumstances in a short time, even clinicians may forget to consider important diagnostic possibilities during workup. The consequences of such errors may range from nothing to life-threatening complications. Isabel is best used as a checklist after the primary decision making has been performed.

Although there is potential concern that users might over-investigate their patients on the basis of its advice, we have not demonstrated this in two large studies (one simulated and the other clinical). Users seem to be able to distinguish between the 'red herrings' and the really important suggestions without much difficulty. It may be thatIsabel will help reduce the number of inappropriate tests done as part of a lengthy diagnostic workup constructed without the help of a checklist such as Isabel.

A dedicated Isabel Medical Content Team is involved with the creation and maintenance of its knowledge base. This team is composed of a number of clinicians. Since it is delivered on the Web, the system is updated almost daily to a small extent. Major updates are performed every 3-6 months.

Isabel is easy to use and provides rapid advice within seconds. In our studies, users took less than 1 minute each to process the diagnostic suggestions and make meaningful clinical decisions. No formal training is required since it is browser-based, and the only infrastructure necessary is an Internet connection.

Isabel is formatted to fit a PDA screen as long as you have a PDA connected to the Internet. Isabel is not downloadable onto a PDA. From your PDA, enter us.isabelhealthcare.com/pda and you will automatically be directed to the pda site.

Some of the early expert diagnosis decision support systems have undergone accuracy studies. These were either conducted by their developers or by independent researchers ( Berner ES et al. NEJM 1994 330(25):1792-6 ). In a study by Graber & Vanscoy "the final ED diagnosis was found in the differential diagnosis generated by Iliad and QMR 72% and 52% of the time respectively." PKC handles one chief complaint at a time and is designed for data to be entered by the patient. A recent study ( a randomized outpatient trial of a decision-support information technology tool ) on PKC showed that "average Coupler session took approximately 18 minutes of employee time to coordinate."

[i] Sandeep B. Bavdekar and Mandar Pawar. Evaluation of an Internet-Delivered Pediatric Diagnosis Support System (Isabel) in a Tertiary Care Center in India :

The overall sensitivity of 80.5% demonstrated in the study is higher than that reported for computer-based diagnostic systems studied by Berner et al. and Graber et al.

[ii] Graber M, Mathews A. Performance of a web-based clinical diagnosis support system using whole text data entry. VA Medical Center, Northport, NY and the Department of Medicine, SUNY at Stony Brook, NY :

In November 2005 Graber found that when key clinical features from 50 challenging CPC cases, reported in the New England Journal of Medicine, were entered into Isabel, the system provided the final diagnosis in 48 cases (96%).

[iii] Borowitz SM, Amy LR, Lyman JA et al. Impact of a Web-based Diagnosis Reminder System on Errors of Diagnosis. Dept of Pediatrics, University of Virginia, Charlottesville, VA and Dept of Health Evaluation Sciences, University of Virginia, Charlottesville, VA.

A study by the University of Virginia in December 2005 found that in 10% of cases, a diagnosis reminder system caused the user to consider a major diagnosis they had not considered and concluded that web-based diagnosis reminder systems could reduce diagnostic omissions and the number of medical errors.

The Isabel diagnosis reminder system has been checked for diagnostic accuracy in both paediatrics and emergency medicine patients. In a large paediatric validation study done in 2000-01, clinical features from 99 hypothetical and 100 real life cases were entered into the Isabel system. The system displayed the 'correct' diagnosis in 95% of the cases. In a more recent evaluation in adult emergency medicine patients, Isabel displayed the discharge diagnosis in >90% of the cases.

Isabel 's impact on decision making has been examined by its developers in both an experimental setting and in a clinical naturalistic environment. In the clinical study, junior clinicians used Isabel in real time and made decisions that impacted on patients. Both studies assessed the benefits as well as the risks of the system advice.

In a quasi-experimental study, subjects of different grades assessed simulated cases representing acute paediatric scenarios, both before and after Isabel consultation. Isabel reminded subjects to consider at least one clinically important diagnosis in 1 in 8 case episodes, and prompted them to order an important test in 1 in 10 case episodes. No clinically significant diagnoses were deleted after consultation indicating that consultation did not result in any deleterious effects.
See peer-reviewed article published in BMC Medical Informatics and Decision Making

In a clinical study at 4 NHS hospitals, junior clinicians used the Isabel system when they needed diagnostic assistance during acute paediatric assessments. They recorded their initial (unassisted) diagnostic workup prior to being provided Isabel advice for the case. Post consultation diagnostic workup was also recorded. Incomplete workups were identified in nearly 45% of the cases by the expert panel. In a third of these, Isabel prompted the consideration of appropriate diagnoses and led to a complete workup. Isabel thus produced a meaningful change in the quality of diagnostic decision making in 14/104 (13.5%) cases. No significant diagnoses were deleted from the workup after consulting the diagnostic aid.
See report submitted to funding body: Department of Health

Isabel’s ability to handle natural language / free text makes interfacing with EMR very simple and straightforward. As far as we know Isabel is the only diagnosis decision support system to be interfaced with EMR. Pre-assigned fields (age, gender, presenting/chief complaints, positive findings) in the EMR are defined. Data from these pre-assigned fields are submitted to Isabel (single click on an Isabel button in the EMR) using a secure interface.

Isabel has been successfully interfaced with high profile US EMR systems.The best way forward would be to have diagnosis decision support systems sitting unobtrusively 'behind the dashboard of the EMR'. Providers seeking diagnosis support need to click just one button. This has already been achieved - Isabel has been interfaced with NextGen, PatientKeeper and A4 Health Systems and we are in discussions with several hospital based EMR vendors. The interface extracts the data that sits in pre-assigned fields - chief complaints, problem lists, positive clinical features from HPI and assessment and of course the age, gender, etc. The next stage should involve extraction of lab data and prescribed drug list.

Using pattern matching software, the Isabel Clinical Alert and Monitoring system (ICAM) allow healthcare professionals in a hospital or group of hospitals to view, in real time, clinical queries entered into an EMR system and diagnoses being considered. Should a provider enter a query similar to another query entered, these systems will generate an alert of this emerging paradigm.

For clinical features entered Isabel provides not just likely diseases but also causative drugs (drugs that might cause the clinical features). Isabel searches Martindale’s 4000 drugs data cross-section ally.

Isabel includes 11000 diagnoses including Bioterrorism diagnoses and can play an integral role in the event of a bio-terrorist attack. Providers understandably lack the necessary clinicalraining and experience to recognize and diagnose bioterrorism conditions. Isabel is able to bridge this skill gap by constructing a list of likely causative bio-terrorist agents like nerve and chemical agents, biotoxins and emerging infectious diseases for a given set of symptoms. You might be interested in this study of older rules based diagnosis decision support system evaluated in terms of bioterrorism response
http://www.cdc.gov/ncidod/EID/vol10no1/pdfs/03-0243.pdf

Most medical error studies find 10-30% of errors are errors in diagnosis. A review of 53 autopsy studies found an average rate of 24% missed diagnoses. ‘ Diagnosing Diagnosis errors: Lessons from a Multi-institutional Collaborative Project’ Gordon D. Schiff MD.Cook County John H. Stroger Hospital and Bureau of Health Services, Chicago, USA. In Advances in Patient Safety - 2. 2005.

Survey of Medical mistakes in the USA by YouGov Nov 2005. This survey of 2201 respondents was commissioned by the Isabel Medical Foundation and showed that 35% of people have experience a medical mistake over the last 5 years. 50% of the mistakes were misdiagnosis, 24% were medication and 18% were operative. This means that 1 in 6 of the US adult population has experience of misdiagnosis and is the same as the figure obtained by the NPSF survey carried out by Harris in 1997.

Isabel is trying to address the issues highlighted by Mark Graber's article Diagnostic Error in Internal Medicine which found that cognitive factors contributed to diagnostic error in 74% of cases studied. The most common cognitive problems involved faulty synthesis. Premature closure, i.e. the failure to continue considering reasonable alternatives after an initial diagnosis was reached, was the single most common cause.

Elstein has suggested the value of compiling a complete differential diagnosis to combat the tendency to premature closure. Arthur S Elstein, Alan Schwarz, Clinical problem solving and diagnostic decision making: selective review of the cognitive literature BMJ 2002;324:729-732.

Residents & faculty (correctly diagnosed 44% and 50% of difficult cases, respectively) were overconfident, placing credence in a diagnosis that was in fact incorrect, in 15% and 12% of cases. ‘Do Physicians Know When Their Diagnoses Are Correct? Implications for Decision Support and Error Reduction’. Charles P. Friedman, PhD et al J Gen Intern Med 2005; 20(4):334-9.