Using the postal code and address information from the APPROACH database, we were able to determine the spatial location of the street address and postal code and the difference in distance between those paired points. The major finding of the present study is that the postal code location closely approximates the location of residence for a large majority of patients with over 80 percent of the postal codes being within 200 meters of the address, and approximately 95 percent of the postal codes being with 500 meters of the street address.
How close is close enough for studies in epidemiology and medical geography? The answer to this question depends on the amount of accuracy needed to answer a specific research question, which in turn depends on the nature of the research question itself. For example, investigation of a regional outbreak of E. coli gastrointestinal disease would probably not be hampered by a random misclassification of locations of 200 meters or less for the majority of cases. In contrast, the planning of fire station or emergency medical service (EMS) networks within city limits is an example of a scenario where the accuracy of postal code locations demonstrated in our study may not suffice. Indeed, dealing with only approximately 80 percent accuracy rates for placing residences within a city block might lead to problems if one were to rely on postal codes only in the planning of locations for EMS and fire stations. The consequence of using postal codes would potentially include misclassification of events and eventually sub-optimal response times and access to service. Accordingly, researchers and policy makers need to consider carefully the geographical issue that they are studying and should use postal codes only if an approximate localization of individuals will suffice. For our research on localization of patients undergoing cardiac catheterization to assess overall geographic equity and access within the City of Calgary, we conclude that the spatial relationships derived from postal codes (and presented in Figure 1) represent a reasonable proxy for the true location of the residence. We are certain, however, that there are geographical scenarios for which more accurate determinations of location are required.
While a large majority of postal code locations were within 500 meters of true location of residence, it is notable that some postal code locations were large distances from the actual street address location (up to 4,482 straight-line distance meters away). The reasons for these highly discrepant cases are unclear, but could include errors in the postal code conversion file geographical coordinates (something that has been anecdotally reported by researchers who work with this file [1, 2], and ), errors in the geocoding process in ArcView 3.2 for the street address, or a missed error in the cleaning process of the APPROACH database. Of course, it is possible that some of these larger differences represent true differences between street address and postal code locations.
The major limitation of the present study is that it focuses only on a single large urban Canadian city. Accordingly, the results do not necessarily apply to other Canadian cities or to cities in other countries. However, there are many reasons to suspect that the findings for the single city studied apply to other Canadian cities, and perhaps also to urban settings in other countries. In contrast, however, the results are unlikely to apply to rural postal codes where single postal codes cover large geographic areas. It is thus clear that additional research is needed to address the locational accuracy of postal codes in other environments. A second caveat to our findings is that we consider address locations geocoded from street network files as the 'reference standard' against which postal code location is compared. Such address locations are, in fact, themselves merely an estimate of location that may or may not be entirely accurate. Of relevance in this regard, Bonner and colleagues  recently compared geocoded address locations with true 'gold standard' location measures derived from Global Positioning System satellite receivers, and demonstrated slight discrepancies between the two. Another important caveat to our findings is that we made corrections to the discrepant address and postal code pairings whenever possible. The result of these corrections is likely to be an overall improvement in accuracy of location that would not otherwise occur if real addresses had not been available. This means that our findings from APPROACH may slightly overestimate the accuracy of postal code locations in 'typical' health databases. A final caveat (that may bias findings in the opposite direction) is that we relied on the Statistics Canada Postal Code Conversion File to determine the latitude and longitude of postal code locations. This is a widely used approach. However, anecdotal reports suggest that the latitude and longitude values in Statistics Canada data may be somewhat erroneous due to a geographical projection problem in the Statistics Canada street network file. These projection problems may in turn adversely affect the accuracy of postal code locations.
Despite these limitations and caveats, our study is informative. Our findings indicate that, although postal code location is not a perfect representation of street address location, the estimate is very close for a majority of cases. Researchers and policymakers interested in conducting and interpreting results of epidemiological or geographical studies need to consider carefully, on a case-by-case basis, whether a misplacement of 200 to 300 meters (or more) in spatial location is problematic to the objectives of their analysis. If the misplacement is not a major concern and given the random nature of that misplacement, the postal code can be used for analysis. In other instances, more precise information on address location should be obtained and used in analysis.