While a number of studies have quantified the occurrence of adverse events in health systems [1, 2, 8, 15, 16], none have examined the impact of these events in a population-based sample of the US population. Using eight years of cross-sectional data from an on-going, annual population-based survey of the US adult population, we have attempted to measure the burden of morbidity arising from these events and to determine the relationship between complications of medical care and socio-demographic variables.
Our estimates of prevalence differ from those conducted by other studies. Extrapolations to the US general population of data from HMPS [1, 2] conducted in New York State in 1984 have indicated that over a million hospitalized patients suffered an injury due to medical treatment [9, 16]. Using similar methodology, a more recent study made similar extrapolations from data collected in Utah and Colorado, arriving at a total population figure slightly less than one million [5]. Even when limited to adults, our results point to population numbers averaging about 1 million over the seven-year period, with a suggestion of a generally increasing trend to values above this level.
Our findings are strengthened by the use of data from subjects randomly selected from the total US population over eight years. Previous research has used hospital-based data to reach conclusions about the experience of the total population. The external validity of such inferences are conditioned on the representativeness of the study population on which the results are based. Hospitals often lack clear indications of the catchment populations to which such generalizations are to be made. This progressive hierarchical leap is hard to justify, especially if primary data from the total population of interest is publicly available, as is the NHIS. Furthermore, our data suggest that a statistically significant relationship exists between such injuries and certain US geographical regions, implying that estimates derived from localities may not correctly reflect the total population's experience.
The NHIS lacks both the ability to externally validate the veracity of self-reported claims of complications of medical care and the specificity of classification offered by hospital-based studies. For instance, one cannot measure the extent to which subjects reported conditions that were due to some disease process instead of a particular complication of medical care. In some cases, medical complications are an unavoidable outcome of therapy. In spite of this, our data captures a feature of inquiry that is missed by hospital-based studies: population estimates include injuries that arise from community-based sources. Thus, two different aspects of the same question are addressed.
A recent Australian study attempted to measure such incidents arising from the general practice setting. Of about 1,500 reports of adverse events received, 44% were due to premature or inappropriate use and 26% reported problems occurring during therapeutic use. About 15% of adverse events were due to the use of contraindicated medication and 11 percent were due to unintended medications or use that was not medically indicated at the time [17]. It has been estimated that such incidents give rise to comparable costs to the health system as that due to all other forms of injury combined (including suicide, falls, homicide, etc.) [18].
In this study, injuries arising from complications of medical care were self-reported. This dependence on the subject's self-reported recall of events is problematic only if differential recall is related to the outcome of interest. However, due to a lack of validation from external sources (i.e., medical records, case notes, etc.), we are unable to exclude the possibility that point estimates within populations or specific subgroups are misrepresented, especially since some subjects were asked to recall events that took place up to one year from the interview.
Plausible situations arise that might account for differential reporting or misclassification. These might take the form of differences in awareness that injuries may be related to certain medical complications, or local or nationwide publicity related to high-profile cases of complications of medical care being applied to personal situations. We do not have enough data to speculate as to how, or to what extent, these differences might have affected the results. However, biases arising from misclassification in this setting will tend to attenuate any relationships found, since there is no a priori evidence that one categorization was more likely than another. We propose four situations in which misclassification might be plausible.
Firstly, reporting behaviour, as with medication compliance or dietary recall, is a complex activity affected by numerous external factors [19, 20]. Secondly, subjects may attribute outcomes of the disease to outcomes of therapy, or vice-versa. Thirdly, the litigious character of participants in the US health system (in both providers and consumers of medical care) has been widely recognized [21, 22]. Lastly, increased media publicity about the impact of adverse events is known to make evaluation of symptoms difficult [23].
Prevalence data was used to estimate risk, but its interpretation is relevant only insofar as it is related to the associations with relevant covariates or groups of covariates suggested by the statistical models. Any suggestion that these results imply causality is an inappropriate appreciation of the complexity of the subject matter and the inadequacy of the primary design of the NHIS. For instance, the prevalence odds ratio is an unbiased estimate of the incidence odds ratio only if the exposure is unrelated to prognosis (ie., duration of illness is the same for exposed and unexposed groups). Since self-reported, cross-sectional data were used, we have no direct information about the longitudinal features of the condition. Hence, even the most fundamental criterion in arguing for the presence of causation - temporality - is unfulfilled.
When modelling the risk for injuries due to complications of medical care, we found significant interaction between age and race. The effect of age had been previously examined in the HMPS [2]. In that study, subjects above 64 years of age were more than twice as likely to have complication arising from medical care compared to those under 45 years. We hypothesized that some degree of residual confounding was present when broad age ranges were used. Our unadjusted results mirror those of the HMPS findings. However, after adjustment for multiple variables, differences emerged. To our knowledge, this is the first indication of the presence of such an interaction. We explored the possibility that these results were due to some bias in the design of the study. The NHIS is known for the consistency of its data checks and response rates for the analysis period were consistently above 90%. Selection bias is unlikely in this setting.
We recognize the limitation of the racial classifications used by the NHIS, especially in the category labelled "Others". However, the lack of available information for persons that could potentially be included in this category prompted us to keep the three-category coding. The results are admittedly unstable, given that analyses were based on much smaller sample sizes. Future research could examine this issue more finely.
If present, non-differential misclassification will generally tend to attenuate predicted differences between groups. For instance, this attenuation might have been the cause of the non-significant results seen in the older age groups. However, if present, the results for the younger age groups are expected to be underestimates of the true value.
The potential immediate and long-term research and policy implications of these findings are many and have been discussed previously in other fora [[24–28]]. This study lends support for the development of common definitions and systems for the routine collection and analysis of data from complications of medical care. These information systems should not only have the capacity to exploit the inherently hierarchical nature of specific health service boundaries (ie., hospitals within counties within states within regions), but also provide a means of promoting a systematic method to the strengthening of approaches to quality of care within the larger health care community.