|Home | About | Journals | Submit | Contact Us | Français|
Little is known about the relative incidence of serious errors of omission versus errors of commission.
To identify the most common substantive medical errors identified by medical record review.
Retrospective cohort study.
Twelve Veterans Affairs health care systems in 2 regions.
Stratified random sample of 621 patients receiving care over a 2-year period.
Classification of reported quality problems.
Trained physicians reviewed the full inpatient and outpatient record and described quality problems, which were then classified as errors of omission versus commission.
Eighty-two percent of patients had at least 1 error reported over a 13-month period. The average number of errors reported per case was 4.7 (95% confidence intervals [CI]: 4.4, 5.0). Overall, 95.7% (95% CI: 94.9%, 96.4%) of errors were identified as being problems with underuse. Inadequate care for people with chronic illnesses was particularly common. Among errors of omission, obtaining insufficient information from histories and physicals (25.3%), inadequacies in diagnostic testing (33.9%), and patients not receiving needed medications (20.7%) were all common. Out of the 2,917 errors identified, only 27 were rated as being highly serious, and 26 (96%) of these were errors of omission.
While preventing iatrogenic injury resulting from medical errors is a critically important part of quality improvement, we found that the overwhelming majority of substantive medical errors identifiable from the medical record were related to people getting too little medical care, especially for those with chronic medical conditions.
People appear to have a natural tendency to be much more disturbed by injury caused by acts of commission compared with injury caused by acts of omission, a phenomenon often referred to as “omission bias.”1–4 For example, people tend to view a death resulting from a vaccine as much worse than a death resulting from not getting vaccinated.5,6 Perhaps this psychologic phenomenon, in part, underlies why the patient safety movement has so strikingly caught the attention of the public and media.7–12 Patient safety, with its emphasis on iatrogenic injury resulting from medical errors,7–12 has garnered widespread attention that greatly exceeds that generated by traditional, long-standing discussions of quality of care, with its emphasis on optimizing care and outcomes.13–15
Although iatrogenic harm is always distressing, research to date suggests that the majority of such injuries are not preventable, but instead appear to result from inherent risks and limitations of medical care.16–20 In addition, the last 15 to 20 years have witnessed an explosion of valuable medical interventions (e.g., improvements in care for heart disease, HIV, diabetes, depression, hypertension, cancer prevention, etc.),15 and virtually all of these valuable interventions carry a measurable risk of iatrogenic adverse events even when provided under optimal circumstances. Therefore, although the risk of iatrogenic harm resulting from misapplications or mistakes in using the ever-growing number of medical treatments (errors of commission) has increased, so has the potential to cause serious injury and death from inadequate treatment (errors of omission) 15,21–23 (see Fig. 1Table 1).
Although scholarly work on patient safety clearly includes an appreciation for both errors of omission and commission, discussions of safety problems, especially those aimed at the general public, have overwhelmingly highlighted preventable iatrogenic injury.7–12,18 Even in the scholarly patient safety literature, little attention has been given to the relative incidence of errors of omission versus commission.7–12 We can find no rigorous peer evaluations of the relative number of major errors of omission versus errors of commission in a representative sample that evaluated both inpatient and outpatient care. Therefore, we conducted a systematic longitudinal review of medical records to examine this issue.
A stratified random sample of patients who had at least 2 outpatient primary care visits in each of 2 years was drawn from 12 Department of Veterans Affairs (VA) health care systems in 2 regional health care networks (Midwest and Western U.S.) that care for over 400,000 people annually. We oversampled 2 chronic medical conditions (diabetes and COPD) to obtain a minimum of 175 cases for each condition. We reviewed 621 cases, with 10% undergoing blinded, independent duplicate reviews to determine inter-rater reliability. We requested all records for each patient for a 2-year time period, inpatient and outpatient, but evaluation of quality was limited to a 13-month period.
Reviewers were required to be board-certified Internists with current or recent outpatient and inpatient experience as well as being conversant with the principles of evidence-based medicine. For each condition evaluated, reviewers assessed the quality of care related to diagnosis and assessment of the condition's initial presentation, the assessment and monitoring of the course of the condition, the treatment of signs or symptoms, exacerbations or complications, and follow-up. They then provided a rating of the quality of care on a 6-point scale (ranging from “very good” to “very poor”). The structured review process specifically cued them to consider the importance of any overuse, underuse, or misuse quality problems, and reviewers were warned against hindsight bias.22–25 Finally, for any care that they rated as “borderline poor,”“poor,” or “very poor,” the reviewers wrote a description of the quality problem/error.
Reviewers were able to scan available records for the 11 months prior to the study period, but their review was restricted to rating the care provided during the 13-month study period (September 1, 1998 to September 30, 1999). The inter-rater reliability (IRR) for the implicit assessments of overall quality varied for different conditions and for different aspects of care but was consistently as good or better than those reported for other implicit review tools (intra-class correlation ranged from 0.16 to 0.46).19,26,27
After reviewing previously proposed classification systems,7,15,17,20,28–31 we chose 3 overarching domains for classifying errors: (1) overuse/underuse/misuse, (2) care function (i.e., diagnosis, treatment), and (3) care modality (i.e., history, physical exam, laboratory test, medication). Pilot testing found that reviewers' verbatim descriptions could be easily divided into errors of commission versus omission, but that it was often difficult or impossible to distinguish between misuse and overuse based on the reviewers' verbatim comments. Therefore, we limited this assessment to underuse versus overuse/misuse.
Two physician reviewers not involved in the original reviews independently classified discrete error descriptions using the previously described taxonomy (see Appendix, available online). Reviewers were also asked to denote errors that they felt were particularly serious (defined by the degree to which the error put the patient at risk of serious injury).23 When the blinded, independent ratings of the 2 reviewers' disagreed with each other (1,038 [11.7%] out of 8,838 total classification decisions on function, modality, and care problem made on 2,946 discrete error descriptions), a third reviewer (M. H.) reviewed the verbatim response. Her rating settled the tie when she felt that one reviewer's assessment was clearly correct (n=886). In a few instances (n=16), she felt that neither reviewer answer was correct. In such cases, a group review was conducted (usually resulting in a classification of “unclear/insufficient information”). If she felt that either answer was reasonable (i.e., difficult or close calls) (n=136), the disagreement between reviewers was settled by random assignment (to prevent the third reviewer from introducing bias by systematically deciding “close calls”).32 Some verbatim descriptions (n=29) did not identify a medical error and were dropped, leaving 2,917 items for analysis. The reliability for classifying verbatim responses was high: overuse/misuse versus underuse (κ=0.75), care function (κ=0.79), and core modality (κ=0.81).
Analyses primarily involved descriptive statistics. After classifying verbatim descriptions of reviewers' comments, as described above, we calculated the frequencies, percentages, and 95% confidence intervals (CI) for the various types of errors, correcting for clustering (by patient and site) and design effects (over-sampling) using the Stata 8.0 statistical package (StataCorp, College Station, TX).
Of the 621 cases, 82% had at least some care that was considered “borderline poor” or worse (Table 2), with an average of 4.7 (95% CI, 4.4, 5.0) discrete errors identified per case. Of the specific conditions, diabetes care had the highest reported error rate (1.7 per diabetes case) and acute conditions had the lowest (0.6 errors per acute condition case). Of errors that could be classified (36 descriptions were rated “unclear/not classifiable”), 95.7% (95% CI, 94.9% to 96.4%) were classified as underuse and 4.3% (95% CI, 3.6% to 5.1%) as overuse/misuse. For the 4 chronic conditions, 4% of errors were caused by overuse/misuse, whereas 14% of errors for acute condition care were caused by overuse/misuse. However, this higher relative frequency of overuse problems was not caused by a higher absolute frequency of overuse errors for acute condition care, but rather, the care for acute conditions had a much lower frequency of underuse errors. For example, the 621 cases averaged 0.13 overuse/misuse errors per case and 2.85 underuse errors per case related to chronic disease care, compared with 0.09 overuse/misuse errors per case and 0.54 underuse errors per case for acute care in the 409 cases with an acute condition.
With regard to errors of omission, over half the problems identified were classified as related to diagnosis and monitoring, with about one third related to treatment (Table 3). Obtaining insufficient information from histories and physicals (25.3%), inadequacies in diagnostic testing (33.9%), and patient not receiving needed medications (20.7%) were all commonly reported problems. Examples of insufficient information from histories and physicals include a patient with facial weakness without further documented neurologic examination and a patient with lower extremity edema and no cardiac examination recorded. Examples of inadequacies in diagnostic testing include a patient with known coronary artery disease and new onset chest pain who received no stress testing, and a patient who presented with night sweats but received no chest x-ray or ppd. Some examples of not providing needed medications include a patient with a previous lacunar stroke who was not on an anti-platelet agent, and a patient with type 2 diabetes who had no medication changes despite marked and repeated elevations of blood pressure. Additional examples can be found in the Appendix.
We were concerned that although underuse errors may be much more common overall, perhaps overuse/misuse errors make up a disproportionate percentage of the most serious medical errors. However, this was not the case. The reviewers only rated 27 of the 2,917 identified errors as being particularly serious, and only 1 of these errors (3.7%) was an error of commission, which involved giving simvastatin to a patient with a reported allergy to statins. Of the remaining 26 major errors, multiple instances were reported for inadequate assessment for or treatment of coronary artery disease (n=14), marked hypertension (n=3), thrombotic risk (n=3), cardiac arrhythmias (n=2), and hyperkalemia (n=2).
A dedication to safety is a critically important responsibility of any industry.7 However, when providing air transportation or selling kitchen appliances, consumer safety is generally limited to considerations of the risks related to using the service or product, whereas, improving consumer safety in medicine requires assessing, and often balancing, the potential harm of inappropriate non-intervention (errors of omission), as well as inappropriate or unsafe intervention (errors of commission) (Fig. 1). In this way, medicine has more in common with the military or law enforcement, in that the risks of friendly fire and intervening too early or too much must be weighed against the risks of enemy fire and intervening too slowly or insufficiently.
Although iatrogenic adverse events are sometimes talked about as if they are synonymous with medical errors, increasing a patient's risk of iatrogenic adverse events will often be the right thing to do.21–23 For example, underuse of thrombolytics, anti-platelet agents, and anticoagulants are common causes of preventable morbidity and mortality.15,33,34 However, appropriate use of these medications clearly and substantially increases iatrogenic adverse events even when done optimally. Of course, misapplications or mishaps in the use of medical interventions can result in iatrogenic injuries that are truly appalling.7,12 Therefore, discussions and considerations of adverse events, patient safety, and medical errors must always consider the appropriateness of medical intervention at the time the decision was made,23,35 to try and avoid second-guessing influenced by hindsight and omission bias.25
To our knowledge, our study is the first systematic examination of the balance between substantive errors of omission versus commission using a detailed peer evaluation of both the inpatient and outpatient medical record. We found that the vast majority of medical errors noted by trained reviewers were related to errors of omission. This finding was consistent across a variety of conditions and was equally true when limiting review to those errors that were felt to be the most serious. Many experts have decried the inadequacies of our health care system for taking care of those with chronic illnesses,15,36 and insufficient care for people with chronic illness was particularly prominent in our study. For example, there were over 3 times more substantive errors of omission related to diabetes care than there were errors of commission related to all conditions combined in the entire patient sample.
Some previous research has examined errors of omission and commission in the inpatient setting17,20,29 or using explicit review criteria.31,37 Although anecdotes of mishaps causing iatrogenic complications tend to be highlighted in most discussions of medical errors, a careful reading of past studies reveals that our results are not that dissimilar from results found in inpatient settings. Evaluations of inpatient errors have consistently found that at least half of major adverse events reportedly resulting from medical errors were related to underuse.17,20,29 For example, the Utah–Colorado Study (UTCOS) found that for inpatient care, excluding adverse events related to neonatal care and patient falls, permanent disability and death were almost 3 times more likely to be caused by underuse errors than overuse/misuse errors.17 Still, our study found an even greater preponderance of underuse errors than previous studies, and we believe that there are several potential reasons for this finding. First, most previous studies mainly or solely examined errors in hospital care, where errors of commission might be relatively more common.17,20,29 Second, most previous studies sampled adverse events and then worked backward to find errors. Such retrospective, unblinded reviews may overestimate errors of commission because even experts are highly subject to hindsight and omission bias.1–5,24,38 On the other hand, it also seems likely that the occurrence of adverse events might increase documentation of otherwise undetectable overuse/misuse errors. Third, the number of errors of omission may have increased since the studies conducted in the 1980s and the early 1990s. As the number of treatments known to produce benefit has increased and cost containment efforts have favored less use, the percentage of errors resulting from undertreatment may also have increased. For example, using an explicit review instrument, McGlynn et al.31 found that patients failed to receive nearly 50% of indicated care.
Our findings are limited to 12 VA health care facilities, and therefore our estimates of underuse versus overuse/misuse should not be overly generalized to other settings. However, we think it likely that the general finding is not limited to VA facilities. Indeed, both overall quality and receipt of recommended care have generally been reported to be as good or better in VA than in other sectors of the US health care system.39–41 A more major and inherent limitation of our study is its reliance on the medical record. Medical errors are frequently not identifiable from the medical record;18,42,43 however, most of the albeit limited information on this topic has not found major differences in the types of errors found in the medical record versus patient or medical personnel reports.18,42,44 In addition, the VA medical record system has been found to be fairly reliable in accessing quality.45 Still, it may be that overuse/misuse errors are much more difficult to detect by medical record review than underuse errors and quality related to technical skill (such as how well a procedure or surgery is performed) is virtually impossible to glean from the medical record. Therefore, our results must be qualified as only pertaining to the types of medical errors that can be detected from medical record review.
What are the implications of our findings? These results clearly show that the attention given to patient safety is well deserved, but also suggests that some portrayals of the nature of the problem may be misleading. A detailed understanding of the overall incidence and importance of errors of omission versus commission can be critical when trying to improve the effectiveness, efficiency, and safety of health care.23,46–48 Many of the system solutions being pursued by patient safety and human factors experts7,15 clearly have the promise of substantially reducing both types of errors and our findings create a stronger case for such safety systems. For example, improving communication between providers, enhancing information availability, and the use of electronic alerts, reminders, and order entry7,15,22 all have direct relevance to reducing errors of commission and errors of omission. However, we risk overlooking important opportunities for designing these systems to optimally improve patient safety if we fail to understand the incidence and full importance of errors of omission.
Further, some proposed safety interventions either directly or indirectly present risks of increasing underuse (such as redundancy systems, compulsory order renewals, mandatory referrals, drug interaction alerts, and pre-authorization). In such instances, our appropriate abhorrence for iatrogenic injuries can lead us to bad decisions if we fail to reflect upon the incidence and relative importance of errors of omission versus commission for the specific situation under consideration. This is another instance in which health care can learn from safety efforts in other industries.49–52 For example, when 4 people died in a British rail accident in 2000, the rail authority reduced maximum allowable rail speeds, which was felt to subsequently contribute to substantial reductions in rail travel—resulting in increased auto deaths, economic losses, and pollution.52 Similarly, an intemperate response to tragic iatrogenic blunders can do more harm than good or result in the diversion of scarce resources from more important, and fixable, health care problems. For example, tunnel vision in trying to eliminate drug-drug interactions could lead to “safety systems” that inadvertently increase patient injury. One of us (R. H.) recently received a pharmacy alert warning that one of his patients was on both aspirin and lisinopril (which have a known drug–drug interaction). However, the risk of aspirin–ACE–inhibitor interactions (though possible with high doses of salicylates) is trivial compared to the thousands of major adverse events and deaths occurring each year resulting from underuse of these medications.15,34,53–55 Even very small reductions in the use of these life-saving medications in response to such well-meaning alerts would far outweigh any potential benefits.
In summary, we found that over 95% of medical errors identified in a review of patients' medical records were caused by patients receiving too little medical care. These results are limited to errors that can be detected by medical record review; nonetheless, they clearly demonstrate that patient safety and quality improvement efforts must carefully reflect upon transgressions related to those things we do and those things we fail to do. Although deserved attention has been directed to iatrogenic harm resulting from medical blunders, research to date suggests that underuse of needed treatments is probably the greatest risk to patient safety.
The work was supported by the Department of Veterans Affairs Health Services Research & Development (HSR&D) Grant IIR 98–103, with additional support from the VA Quality Enhancement Research Initiative (HSR&D DIB #98–001) and the Michigan Diabetes Research & Training Center (NIDDK P60-972573). Drs. Kerr and Asch were supported by Career Development Awards from the HSR&D Office of the Department of Veterans Affairs.
The following supplementary material for this article is available online:
Appendix A. Reviewer Training & Instruction. Appendix B. Random Sample of Overuse/Misuse andUnderuse errors.