In this study, a subset of callers to a regional PCC reflected community pharmaceutical poisoning morbidity as measured by ED visits. Pharmaceutical exposures from an easily identifiable subset of poison control callers were significantly associated with the frequency of poisonings by these same drugs in patients treated in the ED. Significant associations in the frequency of exposures between the UPCC and the ED existed for nine of the top 10 types of drugs responsible for ED visits due to poisoning. Our results suggest that PCC exposure data can be used to construct a real-time pharmaceutical poisoning surveillance system.
Poison control data are attractive as a potential surveillance system due to the large volume of available data and the timeliness of the collection system. The use of trained poison information specialists to document calls presents an opportunity for high specificity and granularity within the data. However, the utility of the NPDS has been questioned because of perceived data limitations.
Previous research examining the relationship of PCC data to community data suggests that poison control exposure records do not correlate well with other indicators of pharmaceutical poisonings.12
However, unlike our study, prior studies only examined fatal poisonings. Surveillance of pharmaceutical poisonings using only death registries may misrepresent community morbidity, since only a small fraction of poisoning patients die from their exposure. Surveillance using death registries alone also shifts the focus away from substances that contribute substantially to the morbidity of poisoning but are rarely fatal.
In contrast, we found that a selection of UPCC data reflected ED morbidity for the most common pharmaceutical poisonings. We found significant associations between nine of the 10 substance categories examined. One of these categories, opiates, has been largely responsible for the increase in fatalities from prescription drug abuse.24
In our analysis, a strong positive association was seen in the frequency of poisonings from opiates between the PCC and the ED. Exposures from another drug category prone to abuse, benzodiazepines,25
were also strongly correlated. The only category not to display significant correlation was salicylates. This category did not undergo a large change in the frequency of exposures and therefore may not have contained enough variability for statistical detection.
One concern of using PCC data for surveillance is that the high volume of low-toxicity, low-morbidity calls15
could influence a surveillance system's ability to detect clinically important poisoning trends. To screen out these calls, we explored three UPCC caller subsets. The UPCC caller subset that was referred for medical evaluation provided both similar demographic and exposure characteristics to the ED patients and enough data for a substance category comparison on a statewide basis. This ‘high-risk’ group was readily identified from the UPCC data fields; thus, this group's exposure information could be easily selected for use in a surveillance system.
One advantage of PCC data is their potential sensitivity to regional trends. Existing sample-based datasets, such as the National Hospital Ambulatory Medical Care Survey and National Health Interview Survey, could be used to examine poisoning trends over time.2–4
However, these systems cannot detect regional changes or changes within individual categories of substances. Much of the responsibility for developing public health policy and legislation to prevent poisonings is at the state and local levels making these systems less helpful. The problem of prescription drug abuse varies widely between states and regions.27
A system capable of identifying regional trends would have key advantages in influencing policy. In our study, we were able to show that associations can be made for individual substance categories in a state with a population of 2.2 million people.29
As of 2006, the entire population of the USA is served by regional PCCs submitting data to the NPDS,15
making this a potentially comprehensive surveillance system for the USA.
According to the Centers for Disease Control and Prevention's Framework for Program Evaluation in Public Health, there are nine key attributes to be considered when evaluating a surveillance system: simplicity, flexibility, data quality, acceptability, sensitivity, positive predictive value, representativeness, timeliness, and stability.30
PCC exposure data have inherent advantages over current sample-based and administrative systems in terms of three of these attributes: timeliness, flexibility, and acceptability. While the system is complex due to its nationwide scope and mission, its fundamental structure and function are easily understandable. To date, most questions about using the NPDS for pharmaceutical poisoning surveillance have centered on concerns about sensitivity, positive predictive value, and representativeness. Our study makes a significant contribution toward addressing these issues.
Limitations and strengths
There are several limitations to this study. The primary limitation was the low granularity and lack of specificity of the ICD-9 coding system for pharmaceutical exposures. Approximately 12.6% of ED visits were coded as ‘unspecified poisoning.’ These visits likely contained some exposures that could have been included in our substance categories. While the poor specificity of the ICD-9 coding system was the primary limiting factor in the granularity of our analysis, limitations of the AAPCC coding schema also exist. Unlike the ICD-9 system, the AAPCC coding schema has not been widely validated. While the poison control data have the advantage that exposures can be classified very finely using POISINDEX codes (Thompson/Micromedex), many exposures are recorded using the more generic AAPCC codes. These generic codes are not uniformly based on ingredient-level data and often represent entire classes of medications; as a result, some information available to the PCC specialist is lost. To be used effectively as part of a surveillance system, the AAPCC coding system would need to permit ingredient-level coding for substances, even when the manufacturer and formulation are unknown. Ideally, PCCs and EDs would code exposures based on a multilevel pharmaceutical taxonomy that permits use of granular ingredient-level data when available but is capable of accepting less granular information as needed. To minimize the loss of information in this process and promote data comparisons, a robust pharmaceutical ontology linked to a national medication standard such as RxNorm would be the most effective.31
We attempted to overcome some of these limitations by using an assignment algorithm augmented with information gained from a probabilistically linked dataset. While the majority of AAPCC codes were assigned using name similarity and ICD-9 poisoning tables, a small percentage of AAPCC codes required information from linkage. Despite the advantages of this technique, in some cases AAPCC codes and ICD-9 codes represented imperfectly overlapping sets of drugs, and some misclassification may have occurred.
Another potential limitation is that substance categories which have stable underlying community poisoning rates would be expected to have inadequate variability to detect an association using our analysis technique. While this problem is a limitation of our study, it should not limit the ability of a surveillance system to detect significant changes in exposure frequency.
These data represent only one state which potentially limits generalizability. Most poison control centers have adopted a uniform software system for assisting with the classification of poisonings; however, regional differences in PCC coding practices may exist.32
Any surveillance system developed would need to keep these possible regional differences in mind.
Finally, fatalities are likely under-represented in these data. While there is substantial overlap between substances causing morbidity and substances causing mortality, there are also significant differences. A surveillance system that combines both morbidity and mortality data would be stronger.
Strengths of this study include the utilization of population-based statewide datasets to compare pharmaceutical poisonings between a regional PCC and EDs. This is the first population-based study to evaluate the association between emergency department exposures and calls to a PCC. Our ability to detect significant associations at the state level suggests the feasibility of regional substance-specific trending.
Second, we developed an empirical system for mapping AAPCC codes to ICD-9 codes in which difficult-to-assign codes were resolved using probabilistic linkage and a defined algorithm. By including in our analysis the 10 most common exposures seen in the ED instead of an individually selected set of substances, we evaluated a representative picture of clinically important poisonings.
Finally, we identified a subset of poison control callers that would be potentially useful in the construction of a surveillance system by excluding many exposures that rarely contribute to community morbidity.