Diagnostic Errors in the ICU

Hospitalized patients continue to be harmed by adverse events and medical errors (Leape. Clin Chim Acta. 2009;404[1]:2). The contribution to errors by preventable diagnostic errors is underappreciated (Newman-Toker et al. JAMA. 2009;301[10]:1060). Studies suggest that the ICU is a particularly highrisk place for diagnostic errors (Shojania et al. JAMA. 2003;289[21]:2849), given the ICU’s high stress, fast pace, and intense environment.

What Constitutes a Diagnostic Error?
Diagnostic errors occur when diagnoses are wrong, delayed, or missed (Combes et al. Arch Intern Med. 2004;164[4]:389; Roosen et al. Mayo Clin Proc. 2000;75[6]: 562; Pastores et al. Crit Care. 2007; 11[2]:R48; Maris et al. Virchows Arch. 2007;450[3]:329). The error may be discovered in the course of care as new data become available or when the current diagnosis is recognized to be incorrect. However, diagnostic errors are often only revealed by the definitive result of autopsy (Graber et al. Arch Intern Med. 2005;165[13]:1493; Pastores et al. Crit Care. 2007;11[2]:R48), when it is too late to effect therapy. Life-threatening diseases that go unrecognized and, therefore, untreated, are perhaps the most concerning of these possibilities. However, incorrect diagnoses resulting in unnecessary testing or inappropriate therapy that confers risk, but little or no benefit, may be more ubiquitous (Newman-Toker et al. JAMA. 2009;301[10]: 1060). Misdiagnoses are usually classified based on their clinical relevance and potential for therapy to have prevented harm. The Goldman (Goldman et al. N Engl J Med. 1983;308[17]:1000) or Battle (Battle et al. JAMA. 1987;258[3]: 339) classification systems are typically used when autopsy recognizes the misdiagnosis. Goldman class I errors are major diagnostic errors in which recognition of the underlying condition before death may have led to different therapeutic options and prolonged survival. Goldman class II errors are major diagnostic errors in which treatment antemortem may not have prolonged survival.

Class I errors have been described in as many as 9% of autopsies in hospitalized patients (5% were considered lethal) (Shojania et al. JAMA. 2003;289[21]:2849) and in 6% to 17.5% of ICU patients undergoing postmortem examinations (Maris et al. Virchows Arch. 2007;450[3]:329; Pastores et al. Crit Care. 2007;11[2]:R48). Class II errors range from 8% to 13% in the ICU. After adjusting for diagnostic improvements over time and declining autopsy rates, analysis suggests that 10% of all hospital deaths involve a major diagnostic error, and 1 in 20 hospital deaths involve potentially preventable class I errors, while as many as 1 in 10 ICU deaths has such an error (Shojania et al. JAMA. 2003;289[21]:2849).

Discrepancies found at autopsy create a record of these errors; however, without a definitive test that shows diagnostic errors during life, both lethal and nonlethal diagnostic errors get ”lost in the chart.“ Nonlethal diagnostic errors in the ICU may also affect longterm outcomes, yet this group remains largely undefined and unexplored. An example would be the failure to recognize subclinical status epilepticus that may leave the patient alive but in a persistent vegetative state (Drislane et al. J Clin Neurophysiol. 2008;25[4]:181).

Of course, some diagnostic errors may be completely harmless and others may be caught before harm occurs. However, just because a diagnostic error did not cause harm does not mean that it is acceptable or unimportant. To date, almost nothing is known about these incidental errors, but the process by which they occur could provide us with valuable clues on how to implement system strategies that may identify and reduce the harmful ones. Traditional evaluations of care, such as morbidity and mortality conferences or root cause analysis investigations, rarely address near misses.

A Systems-Based Approach
Can we prevent ICU diagnostic errors and their resultant harm? This is a great challenge because we do not yet have a full perspective on the scope of the problem. With that said, we know what does not work. The culture of the ABCDs (accuse, blame, criticize, deny) that commonly surfaces when diagnostic errors occur is often counterproductive, reinforces a culture of “defensive medicine,” and fails to address the root causes of the errors and the inherent fallibility in the system. We suggest using systems-based solutions for recognizing and reducing diagnostic errors. While diagnostic errors may result from “thought process” breakdowns in providers, other systemoriented factors, such as data and information management, presentation, integration, and communication, may be vastly more important and are ripe for targeting by systems-based principles.

Systems-based approaches include implementation of comprehensive unit safety programs (CUSPs) to effect culture change and adherence to the principles of safe design – standardization, creation of independent checks (tools such as check-lists and staff empowerment encourage staff to speak up when something is not right), and learning from defects – when things go wrong. CUSPs involve all stakeholders (nurses, doctors, administrators, and others) at a local unit level who work proactively to identify risk for patient harm. Such systems interventions have been shown to be very effective at nearly eliminating some adverse events once considered inherent in patients in the ICU (Pronovost et al. J Crit Care. 008;23[2]:207; Berenholtz et al. Crit Care Med. 2004;32[10]:2014).

“Learning from defect” (LFD) strategies, a second systems-based approach, may also be useful in prevention of diagnostic errors. LFD is a proactive root cause analysis–like strategy that emphasizes not only causal system factors for adverse events but additionally seeks to uncover mitigating factors that may be capitalized upon to improve the system itself and reduce future harm. This strategy is local and streamlined to allow individual units to address their local problems (Pronovost et al. Jt Comm J Qual Patient Saf. 2006;32:102).

Creating a Framework for Improvement
What are the root causes of ICU diagnostic errors? We do know that during off-hours, the risk of misdiagnosis goes up (Kollef. Crit Care Med. 1991;19[7]:906; Okello et al. Injury. 2007;38[1]:112); however, other less-described causal factors, whose contributions remain ill-defined, exist. These may include high-complexity illnesses, alarm fatigue, stress, excessive workload (Donchin et al. Curr Opin Crit Care. 2002;8[4]:316), inappropriate staff-to-patient ratios, questionable qualifications of ICU staff physicians (Pronovost et al. JAMA. 2002;288[17]: 2151), and staffing models that do not fit the particular ICU’s needs.

Cognitive and contextual causes that lead to thought-process diagnostic errors may be ascribed to four factors. First, we all have a tendency to focus on the big or pressing problems. The notion that there is a single explanation for a patient’s condition (“Occam’s razor”) doesn’t necessarily apply to patients in the ICU who typically have multiple problems on admission or tend to accumulate them during their stay. Our universal approach to the overwhelming problem risks our missing smaller elements, such as a cervical fracture in a patient with polytrauma ( Janjua et al. J Trauma. 1998;44[6]:1000). Second, critical care physicians are often so busy chasing the “usual suspects” that they can become overly focused. Lack of response to therapy should be considered indicative of a possible misdiagnosis, but patients in the ICU are often so sick that immediate response to treatment is not guaranteed, even with the correct diagnosis. We may become so wedded to the first diagnosis that we cannot easily entertain the possibility that the current diagnosis is incorrect. Third, information is lost to us. We obscure the patient’s ability to participate in care (sedation, restraints, intubation) and relay to us whether he or she is experiencing changing or new symptoms. Additionally, physical examination in the ICU has poor sensitivity and specificity (Crowther et al. Intensive Care Med. 2005;31[1]:48; Drislane et al. J Clin Neurophysiol. 2008;25[4]:181; Hotson et al. Brain. 1976;99[4]:673), and laboratory and imaging studies are difficult to interpret (eg, D-dimer levels) (Crowther et al. Intensive Care Med. 2005;31[1]:48) or cannot be performed due to logistical or medical challenges (eg, MRI in patients with an automatic implantable cardioverter defibrillator). Fourth, in the ICU, as in many areas of medicine, we are limited in our current scientific understanding or by available diagnostic means. Finally, we may have information overload. The ICU is so complex and inundated with alarms and data that we tend to focus on only the highest perceived threats and may ignore other information that is equally important.

Systems-oriented solutions to address these overburdened situations and reduce ICU errors may need to be generic, such as mandatory staffing by critical care–certified ICU physicians (Pronovost et al. JAMA. 2002;288[17]:2151). Others, however, will need to be specific to particular clinical contexts, such as standardization with structured diagnostic algorithms and checklists to ensure cognitive consistency and thoroughness for a particular clinical problem (eg, unexplained hypotension). Additionally, technological solutions to bring order to the chaos of data presentation, integration, and decision analysis in the ICU environment will need to be developed.

Diagnostic errors clearly exist in the ICU, despite the aggressive and organized care that we provide. Diagnostic error type and incidence may vary, but they undoubtedly lead to some level of harm, and all harm should be viewed as a “never should happen event.” However, the incidence and harm of ICU misdiagnosis are difficult to quantify and must be addressed more scientifically. To date, misdiagnosis has received too little attention. It’s time to more fully define this problem and better diagnose and treat diagnostic errors.

Dr Bradford D. Winters, PhD
Departments of Anesthesiology and Critical Care Medicine and Surgery
and
Dr David Newman-Toker, PhD
Department of Neurology
Johns Hopkins School of Medicine
Baltimore, MD