Context. Medical errors are an important problem for hospitalized adult inpatients. However, medical errors in children remain comparatively understudied, and published research has been relatively limited.
Objectives. To investigate the national rates of hospital-reported medical errors in pediatric inpatients over the period 1988–1997; and to determine the association of patient and hospital characteristics with the occurrence of hospital-reported medical errors in children.
Design, Setting, and Patients. A nonconcurrent cohort study of hospitalized nonnewborn pediatric patients in the United States ≤18 years of age. Data from the Healthcare Cost and Utilization Project for the years 1988, 1991, 1994, and 1997 were used for these analyses.
Main Outcome Measure. The occurrence of hospital-reported medical errors.
Results. The national rate of hospital-reported medical errors in hospitalized children ranged from 1.81 to 2.96 per 100 discharges. These medical error rates were statistically lower in 1988, with the years 1991, 1994, and 1997 not being statistically different from each other. There were no consistent differences in the rates of medical errors when stratified by gender, race, payor status, or median household income of the patient’s zip code across years. There was, however, a statistically significant relationship between higher median household income and increasing medical error rates; this trend was consistent across all 4 years. Similarly, children with special medical needs or dependence on a medical technology also had significantly higher rates of hospital-reported medical errors. Although hospital size did not seem to be related to the rate of medical errors, private for-profit hospitals consistently reported lower rates, whereas urban teaching hospitals in all years but 1997 reported higher rates of medical errors.
Conclusions. These data highlight both the strengths and limitations of administrative data in the investigation of medical errors. Substantively, they suggest fruitful areas for additional and more detailed study, notably children with special medical needs.
The Institute of Medicine’s1 recent report on patient safety identified medical errors as a significant contributor to patient morbidity and mortality in adults, with an estimated 44 000 deaths per year caused by medical errors.1,2 In addition, medical errors on a national level are estimated to cost between $17 billion and $29 billion annually in the United States.1
Medical errors and iatrogenesis in children remain comparatively unstudied. As part of a broader study of national pediatric medical care utilization, McCormick et al3 reported that in 1996, 0.8% of all pediatric discharges included a complication of medical care; however, there are no details regarding the characteristics associated with these adverse events. Brennan et al4 in the Harvard Medical Practice Study investigated medical injuries in New York State in 1984 and derived population estimates and rates according to age and sex. Nonnewborn pediatric patients ≤5 years of age had a mean rate of adverse events of 2.7 per 100 discharges.4 In a follow-up study, Leape and colleagues5 further classified these events in this age group as operative and nonoperative. However, neither the rate of such medical errors on a national level nor the risk factors for their occurrence have been described.
The pediatric literature does contain numerous reports of medical errors and medical misadventures in children, which help to identify risk factors for their occurrence.4–7 Kaushal and colleagues8 investigated medication errors and adverse drug events experienced by hospitalized pediatric inpatients in a single, leading teaching institution. Their findings revealed a rate of 55 medication errors per 100 pediatric admissions.6 There is the suggestion that increases in medically complex pediatric care requiring newer technologies may also lead to complications.6–12
If improvements in the safety of medical care for children are to take place, additional research quantifying the incidence of more generally occurring complications and describing the epidemiology of those iatrogenic complications is required. Smith et al described the occurrence of medical errors using administrative data.13 They used the International Classification of Diseases (ICD) codes to calculate the rates of “medical errors.”11 In this system, the authors made a distinction between “medical injuries” and “true injuries”. A medical injury is an adverse effect or a complication of medical or surgical care. “True injuries” represent injuries that occur antecedent to medical care and are classified as traumatic or nontraumatic in nature. This distinction between medical and true injuries represents an organized, coherent, and reproducible framework that we have used to define medical errors in this study of hospitalized children.
The Healthcare Cost and Utilization Project (HCUP) dataset is a multisector collaborative effort that profiles discharges from community hospitals in over 20 states.12 HCUP is the largest and most complete hospital inpatient dataset in the United States12 and is currently available for the years 1988–1997. It contains anonymous information on patients, providers, organizational systems, and resource use.12
In each year the number of hospitals ranges from 759 to over 900 and the total number of inpatient discharges ranges between 5 and 6.5 million. The design of HCUP is based on a stratified probability sample of hospitals with sampling probabilities proportional to the number of hospitals within a given stratum.3,12 The strata are derived from the hospital characteristics of ownership, size, teaching status, region and rural or urban location,12 and the sampling frame consists of all community hospitals from the participating states.3 For the purposes of HCUP, “community hospitals” refers to “all nonfederal, short-term, general and other specialty hospitals.” A 20% stratified sample of such hospitals is obtained with this method. All discharge records for each sampled hospital are included, with selected diagnostic, utilization, and patient information. With the application of proportionate weights, each discharge can be extrapolated to represent a number of hospital discharges within the stratum of hospitals and thus population estimates.
All HCUP discharges of nonnewborn pediatric patients <19 years old occurring in 1988, 1991, 1994, and 1997 have been included in these analyses. The choice to exclude newborns arises from the unique circumstances of this overwhelmingly healthy group that is hospitalized briefly for special circumstances. We also note that because of their disproportionate contribution to all pediatric discharges, newborns are included in HCUP using a different sampling strategy.
Defining Medical Error
The definition used in the description of medical errors among adults and the definition used in analyses reported here relies on the ICD-9 coding of medical errors using diagnosis codes 996–999 plus 995.2.11 The primary outcome measure of this study is an ICD-9 code indicating a hospital-reported medical error. We surveyed each sampled discharge meeting the inclusion criteria for an ICD-9 diagnosis from any of the 15 diagnosis fields indicative of a hospital-reported medical error (ICD-9 996–999, as well as ICD-9 995.2). ICD-9 codes 996–999 are specifically designated to record complications of medical care and are subdivided based on both the context in which they occur (medical vs procedural/surgical) and by their impact on specific organ systems. For example, “mechanical complication of genitourinary device, due to indwelling urethral catheter” (ICD-9 code 996.31) is included as a specific procedural complication.
We have also included selected diagnoses labeled “certain adverse events not elsewhere classified” (ICD-9 codes 995). Although portions of this broad group are clearly not applicable to medical errors (for example, ICD-9 code 995.5 representing “child maltreatment syndromes”), others are. In particular, we included ICD-9 995.2 (unspecified adverse effect of drug, medicinal, or biological substance).
We determined the association of medical errors as defined above with selected patient and hospital characteristics. The patient level variables included age, gender, race, payor, admission source, admission type, the presence of a special medical need or technology as defined by coincident ICD-9 “V-codes,” median household income of the patient’s zip code of residence, length of stay (LOS), discharge disposition, and survival. The hospital characteristics examined include location, bed size, ownership, and teaching status. HCUP defines hospital categories by bedsize, ownership, and teaching status.12 Teaching status is only defined for hospitals in an urban environment.12 For the survival and LOS variables, we also calculated the weighted rate of death among patients with and without medical errors and the mean LOS for all those with and without medical errors.
We derived the estimated numbers and percentages of such errors in the entire country for each of the years using SUDAAN software (Research Triangle Institute, Research Triangle Park, NC). SUDAAN not only generates weighted estimates, but also produces standard errors that incorporate the complexity of the sample design. To examine statistical trends, both over time and within years, we performed weighted least squares regressions. These regression analyses used the inverse of variances calculated by SUDAAN to weight each of the rates. The slope is therefore interpreted as the rate of change either over time or within the ordinal category (eg, LOS). All these weighted regressions were performed in SAS PROC GLM (version 8.1; SAS Institute, Cary, NC).
Table 1 provides the total and category-specific rates of ICD-9 coded medical errors by year. The national rate of hospital-reported medical errors in hospitalized children ranged from 1.81 to 2.96 per 100 discharges. These medical error rates were statistically lower in 1988, with the years 1991, 1994, and 1997 not being statistically different from one other (Table 1). Similarly, procedural complications (ICD-9 996) were statistically more common in 1991–1997 than in 1988. Rates of reported errors attributable to drug errors showed a statistically significant decreasing trend over the 4 years (P = .02), the average annual rate of this decrease was 1 per 100 admissions.
The patient characteristics associated with hospital reported medical errors in children are presented in Table 2. There were no consistent differences in the rates of medical errors when stratified by gender, race, payor status, or median zip code household income across years. There was, however, a statistically significant relationship between gender and medical error rates. Males had significantly higher rates of medical errors than females in all of the years studied. In addition, a relationship between higher median zip code household income and increasing medical errors rate was found; this trend was consistent across all 4 years (P1988 = .01, P1991 = .0001, P1994 = .0001, and P1997 = .0004), and the average rate of this increase ranged from 1.13 (1997) to 1.33 (1994) per 100 admissions. Children in the 6- to 12-year age range experienced the highest rate of medical errors in all of the years studied. Similarly, children with V-code defined special medical needs or who were technology dependent also had significantly higher rates of hospital-reported medical errors. Not unexpectedly, children discharged to routine care (home) had lower rates than those discharged to other facilities. Those pediatric patients admitted from another health care facility, whether a long-term care facility or another hospital, had significantly higher rates of medical errors than those admitted from the hospital’s emergency department or through a routine admission process (Table 2). Across all four years, patients admitted on an urgent or emergent basis appear to have lower medical error rates than those admitted on an elective basis.
Table 3 presents the institution-level characteristics associated with hospital-reported medical errors in children. Although hospital size is not related to the rate of medical errors, private for-profit hospitals consistently report lower rates, whereas urban teaching hospitals in all years but 1997 report higher rates of medical errors.
The outcomes of death and LOS associated with medical errors are presented in Table 4. The data concerning LOS and hospital-reported medical errors are particularly interesting. Patients with hospital-reported medical errors have significantly longer LOS overall, and beyond the shortest LOS (0–1 day) there is a dose-response relationship between LOS and the incidence of such medical errors. Children with medical errors who remained hospitalized for >5 days had a higher rate of errors in all of the years studied. The mean LOS was 2 to 3 times higher for medical error patients than for those patients who did not experience medical errors (Table 4). There was no statistically significant trend of increasing LOS across any of the years studied (P1988 = .08, P1991 = .14, P1994 = .25, and P1997 = .72).
Children with medical errors had significantly higher death rates than children without medical errors. These findings were consistent across all of the years studied (Table 4). In addition, there was a significantly higher rate of death for patients with medical errors in the years 1991, 1994, and 1997 than in 1988, with a statistically significant decreasing rate over time (P = .019) and an average rate of change of 1 per 100 admissions.
These data from a nationally representative dataset provide some of the first data regarding the general problem of hospital-reported medical errors in pediatric inpatients. By using a nationally representative inpatient sample, it is possible to discuss patterns in the broad group of pediatric patients. We now have an understanding that hospital-reported medical error in hospitalized children is a relatively rare event occurring in <3% of hospital discharges. This rate has increased from 1988 to 1991, but remained stable from 1991 to 1997. Furthermore, children with increasingly complex medical care have higher medical error rates, which is consistent with data derived from studies on adult patients. Children with medical errors also had higher associated LOS and mortality rates throughout the years studied. Therefore, this work primarily provides an important reference point for subsequent and more specialized studies of specific patient groups such as children with chronic illnesses who may already be at risk of prolonged hospital stays and mortality.
Administrative datasets provide extensive data on hospital encounters, patient characteristics, organizational structure, and resource utilization associated with each diagnosis. Their analysis has informed the discussion of topics ranging from error surveillance to resource utilization.13,14 Of importance for the work on medical errors, they represent a coherent and reproducible framework that is especially useful for analyzing relatively rare events such as medical errors. Most importantly, the findings of the current study compare favorably to the pediatric error rate of 2.7 per 100 admissions found by the manual review methods used in the Harvard Medical Practice Study.
Although medical errors occur in a variety of settings, hospitals are an environment in which their description is particularly important. Although medical errors recorded as discharge diagnoses could have occurred outside the hospital and led to the admission, available data suggests that this is generally not the case. In a study of medical errors in 15 000 adult discharges, 80% of medical errors were found to have occurred in the hospital rather than in other health care settings.1
There are, however, several important limitations in this study that need to be considered. First, medical errors are undoubtedly underreported in administrative databases.1,8 In addition, these analyses are subject to the criticism that such underreporting may be biased by a variable threshold for recording such events in various institutions or for various types of patients over time, thus affecting any conclusions drawn from our results. Second, the use of administrative data creates additional limitations. Most notably, it is impossible to separate antecedent and consequential events in relationship to medical errors. For example, the patient who is more seriously ill and hospitalized longer has an increased “exposure” to medical treatment and may therefore experience more medical errors. On the other hand, significant medical errors may also lead to the need for more interventions and increased LOS. Unfortunately, administrative databases cannot be used to disentangle the sequence and direction of causality linking these variables. It would be inaccurate to interpret results obtained here as demonstrating a cause and effect relationship. Rather, they can serve only to identify associations.
Third, administrative data do not provide detailed clinical and physiologic information. Ideally, a study of medical errors would be able to describe differences in acuity and clinical status between patients to better identify those factors placing patients at risk for medical errors. Similarly, the definitions for children with chronic conditions and technology dependence are far from ideal as they are based solely on the presence of documented diagnostic comorbidity.
Any single study is limited, whether by the idiosyncratic circumstances of locally collected data, the incompleteness of national data, or by biases imposed by physician review teams.15 There is, however, an urgent need for interventions to reduce and limit medical errors. This descriptive study does not, of course, constitute or even directly lead to, such an intervention. On the other hand, it is our hope that it will contribute further to the descriptive epidemiology of medical errors in the large and potentially vulnerable group of pediatric patients.
This work was supported in part by the Agency for Healthcare Research and Quality (R03-HS11022-01) and by Children’s Research Institute, Research Advisory Council grant 118.
- Received January 8, 2002.
- Accepted July 8, 2002.
- Reprint requests to (A.D.S.) Department of Critical Care Medicine, Children’s National Medical Center, 111 Michigan Ave, NW, Washington, DC 20010. E-mail:
Preliminary results were presented at the Academy for Health Services Research Annual Meeting; June 10–12, 2001; Atlanta, GA; and the Pediatric Academic Societies Annual Meeting; April 28–May 1, 2001; Baltimore, MD.
- ↵Institute of Medicine, Committee on Quality of Health Care in America. In: Kohn LT, Corrigan JM, Donaldson MS, eds. To Err is Human Building a Safer Health System. Washington, DC: National Academy Press; 2000
- ↵Institute of Medicine, Committee on Quality of Health Care in America. Crossing the Quality Chasm: A New Health System for the 21st Century. Washington, DC: National Academy Press; 2001
- ↵Peery CL, Chendrasekhar A, Paradise NF, et al. Missed injuries in pediatric trauma. Am J Surg.1999;65 :1067– 1069
- ↵Barr DP. Hazards of modern diagnosis and therapy—the price we pay. JAMA.1955;159 :452
- ↵Advisory Panel on Research Uses of Administrative Data of the Northwestern University/University of Chicago Joint Center for Poverty Research. In: Hotz VJ, Goerge R, Balzekas J, Margolin F, eds. Administrative Data for Policy Relevant Research: Assessment of Current Utility and Recommendations for Development. Chicago, IL: University Press; 1998
- Copyright © 2003 by the American Academy of Pediatrics