US 20020082805 A1 Abstract A method for measuring a degree of association (
58) between, and for selectively creating a grouping (40) of, n plurality of spatially referenced physical events (22) of a predetermined physical characteristic. The method includes the steps of assembling n plurality of physical events (26), assembling a universe of possible sample locations (36), determining a reference distribution (54), determining a restricted distribution (56), and determining the degree of association (58) between the n plurality of physical events. Specifically, the physical events each have an indicia of location and a physical characteristic above a threshold. The step (54) of determining a reference distribution is conducted by calculating a test statistic (78) for each of n′ plurality of random allocations (74) of the n plurality of physical events over the selected n plurality of sample locations. Further, the step (58) of determining a restricted distribution includes calculating the test statistic (82) for each of n″ plurality of restricted random allocations (80) of the n plurality of physical events over the n plurality of sample locations (62). Claims(19) 1. A method for measuring a degree of association between n plurality of dimensionally referenced physical events of a pre-determined physical characteristic, said n plurality of physical events each having locations of observation, comprising the steps of:
assembling n plurality of physical events each having an indicia of location and a physical characteristic above a first threshold; assembling a universe of possible sample locations; selecting n plurality of sample locations from said universe; determining a reference distribution by calculating a test statistic for each of n′ plurality of random allocations of the n plurality of physical characteristics over said n plurality of sample locations; determining a restricted distribution by calculating said test statistic for each of n″ plurality of restricted random allocations of the n plurality of physical characteristics over said n plurality of sample locations; and determining the degree of association between said n plurality of physical events based upon said determined reference and restricted distributions. 2. The method of 3. The method of a) selecting n plurality of sample locations to include x number of sample locations in said first sub-area and y number of sample locations in said second sub-area;
b) randomly allocating the physical characteristics from said n plurality of physical events over said n plurality of sample locations;
c) calculating said test statistic; and
d) repeating steps (a), (b), and (c) to generate said restricted distribution including n′ calculations of said test statistic.
4. The method of a) selecting n plurality of sample locations to include x number of sample locations in said first sub-area and y number of sample locations in said second sub-area;
b) randomly allocating the events from each of said x number of n plurality of physical events in said first sub-area over said x number of sample locations and randomly allocating the physical characteristics from each of said y number of n plurality of physical events in said second sub-area over said y number of sample locations;
c) calculating said test statistic; and
d) repeating steps (a), (b), and (c) to generate said restricted distribution including n″calculations of said test statistic.
5. The method of selecting a critical value of the reference distribution;
calculating a credibility of the restricted distribution by determining the proportion of the restricted distribution that is as or more extreme than said critical value; and
determining whether said credibility evidences a statistically significant level of association for the physical characteristics.
6. The method of 7. The method of 8. The method of 9. The method of 10. The method of 11. A computer-implemented method for creating a cluster of dimensionally referenced physical events, said cluster representing a dimensional region encompassing n plurality of said physical events exhibiting a predetermined physical characteristic above a predetermined threshold, comprising the steps of:
assembling a first data structure indicative of the locations of observation and the physical characteristic for said n plurality of physical events; assembling a second data structure indicative of a sample space; determining a reference distribution via a computer processor by calculating a test statistic for each of n′ plurality of random allocations of the n plurality of physical events over n plurality of first sample locations, said computer processor communicating with said first and second data structures and configured to select said n plurality of first sample locations from said second data structure; determining a restricted distribution via said computer processor by calculating said test statistic for each of n″ plurality of restricted random allocations of the n plurality of physical events over n plurality of second sample locations, said computer processor selecting said n plurality of second sample locations from said second data structure; determining the degree of association between said n plurality of physical events based upon said determined reference and restricted distributions; and creating a grouping of said n plurality of physical events only when said degree of association exceeds a predetermined value. 12. The method of 13. The method of a) selecting said n plurality of first sample locations to include x number of sample locations in said first sub-zone and y number of sample locations in said second sub-zone;
b) randomly allocating the characteristics from said n plurality of physical events over said n plurality sample locations;
c) calculating said test statistic; and
d) repeating steps (a), (b), and (c) to generate said reference distribution including n′ , calculations of said test statistic.
14. The method of a) selecting said n plurality of second sample locations to include x number of sample locations in said first sub-zone and y number of sample locations in said second sub-zone;
b) randomly allocating the events from each of said x number of n plurality of physical events in said first sub-zone over said x number of sample locations and randomly allocating the events from each of said y number of n plurality of physical events in said second sub-zone over said y number of sample locations;
c) calculating said test statistic; and
d) repeating steps (a), (b), and (c) to generate said restricted distribution including n″ calculations of said test statistic.
15. The method of selecting a critical value of the reference distribution;
calculating a credibility of the restricted distribution by determining the proportion of the restricted distribution that is as or more extreme than said critical value.
16. The method of 17. The method of 18. The method of 19. A method for measuring a degree of association between physical events, comprising the steps of:
assembling a plurality of physical events; assembling a universe of possible sample locations; determining a degree of association between said plurality of physical events by randomizating said physical events over said sample locations. Description [0001] This invention was made with government support by the United States of America under Grant No. R43 CA65366 awarded by the National Cancer Institute. The United States government has certain rights in the invention. [0002] 1. Technical Field [0003] The present invention relates generally to a method for detecting statistically significant dimensional relationships between physical events and, more particularly, to a method and apparatus for measuring a degree of association between spatially referenced physical events in order to group these physical events when appropriate. [0004] 2. Discussion [0005] It is difficult to overstate the importance of accurately measuring localized occurrences of a physical event such as, for example, disease outbreaks or unsafe pollution concentrations. Incorrectly identifying the existence of such an event can lead to unnecessarily alarming individuals in the “affected” area as well as to causing the expenditure of resources, monetary and otherwise, better allocated elsewhere. Potentially even more devastating is the failure to recognize the existence of a localized event as early as possible. Left unchecked, highly contagious diseases can spread to cause an epidemic while environmental conditions such as increased pollution levels can irrevocably damage fragile natural balances. [0006] Physical events are often spatially or temporally related with localized occurrences referred to as clusters. In these instances, the ability to accurately determine the existence of localized occurrences of physical events depends in part upon the specificity of the spatial or temporal property of each event. Despite the need to accurately measure statistically significant clustering in a variety of contexts, currently available modeling techniques do not accurately reflect the location of each event and therefore too often lead to incorrect inferences. This problem is particularly troublesome in spatially referenced physical events having uncertain spatial locations. [0007] Sources of location uncertainty arise in a variety of contexts. For example, uncertainty can arise in an epidemiologic context due to the anonymity commonly maintained during the reporting of health events, the uncertainty of exposure locations given the mobility of human activity, and the transient nature of many environmentally transmitted disease causing agents. Uncertainty is amplified by the recording of event locations based upon zip code zones, census tracts, or grid nodes. Location uncertainty is also prevalent in the analysis of other spatially referenced physical events such as in the environmental and physical sciences (e.g. biology, geology, and hydrology). [0008] Randomization testing of recorded events is commonly used to infer whether a spatial pattern exists within the sample of spatially referenced physical events. In these tests, the statistical significance of the spatial pattern is generally evaluated through the use of actual or estimated sample locations. When the actual locations of the samples are uncertain, a model is used to approximate the locations of the samples. The most frequently used method for approximating the location of a sample is the centroid model which assigns the area centroid location to all cases or samples occurring within an area. [0009] A particular disadvantage of using randomization tests based upon centroid approximations is that the approach does not consider the spatial distribution of the at-risk population. As a result, approximations based upon the centroid of an area rather than the distribution of the at-risk population create an unnecessarily and inaccurately small universe of possible sample locations. For example, in epidemiological analyses, the universe of sample locations for randomization is more properly related to the geographic distribution of the human population in general and, more particularly, to the distribution of individuals at risk for a particular disease. [0010] Additionally, randomization tests are problematic for spatial data because currently used techniques assume that the sampling space consists of the locations at which the observations were made. That is, they erroneously assume that the universe of possible locations consist entirely and solely of the sample locations. However, in most situations, other locations in the study area could have been sampled. As a result, the sampling space for the spatial randomization test is incorrectly specified and the distributions generated during the test pertain only to the sample locations rather than the at-risk population within the study area. This incorrect approximation leads to detection errors and the potentially dire consequences associated therewith, whether the locations of the physical events are certain or uncertain. [0011] Accordingly, it is an object of the present invention to provide a method for accurately determining the degree of association between physical events. [0012] A further object of the present invention is to provide a method for accurately determining the degree of association between physical events having uncertain locations. [0013] Another object of the present invention is to determine the degree of association between a plurality of spatially referenced physical events based upon an analysis of reference and restricted distributions of a test statistic. [0014] Still another object of the present invention is to determine the relative degree of illness for a given area based upon a comparison of the degree of association between the physical events to a threshold value. [0015] A further object of the present invention is to determine the degree of association between a plurality of spatially referenced physical events through the use of a location model that reflects the spatial distribution of an at-risk population. [0016] The present invention provides a method for measuring a degree of association between, and for selectively creating a cluster of, n plurality of spatially referenced physical events of a predetermined physical characteristic. The method includes the steps of assembling n plurality of physical events, assembling a universe of possible sample locations, determining a reference distribution, determining a restricted distribution, and determining the degree of association between the n plurality of physical events. Specifically, the physical events each have an indicia of location and a physical characteristic above a threshold. The step of determining a reference distribution is conducted by calculating a test statistic for each of n′ plurality of random allocations of the n plurality of physical events over the selected n plurality of sample locations. Further, the step of determining a restricted distribution includes calculating the test statistic for each of n″ plurality of restricted random allocations of the n plurality of physical events over the n plurality of sample locations. [0017] Further objects, features and advantages of the present invention will become apparent to those skilled in the art from studying the following detailed description and the accompanying drawings, in which: [0018]FIG. 1 is a schematic illustration of a method according to the present invention shown in relation to the physical events of interest; [0019]FIG. 2 is a flow chart showing steps of the preferred method with reference to a hypothetical spatial area of analysis; and [0020]FIG. 3 is a graphic illustration of an exemplary reference distribution and restricted distribution of the Mantel test statistic. [0021] The following description of a preferred embodiment of the present invention is merely exemplary in nature and is not intended to unduly limit the scope of the claimed invention. Moreover, the following description, while depicting the invention for use with spatially referenced epidemiological events, is intended to adequately teach one skilled in the art to use the method and apparatus to measure a degree of association between spatially referenced physical events, particularly those events having uncertain locations, regardless of the underlying nature or characteristic of the event. Specifically, those skilled in the art will appreciate that the method and apparatus described and claimed herein is applicable to determine the degree of association for a variety of spatially referenced physical events including environmental events related to geology, hydrology, or pollution control. [0022]FIG. 1 generally illustrates the implementation of the method and apparatus of the present invention [0023]FIG. 1 further illustrates a structure for implementing the present invention as well as the relationship of this structure to physical area [0024] With continued reference to FIG. 1, the physical events [0025] In sum, the present invention determines a degree of association between events that is representative of the actual association between the events. While the events have an actual degree of association based upon the proximity of the event locations, the actual association is incapable of measurement because the events are reported with locations that are generally not reflective of the actual event locations. The determined degree of association is used to selectively create an event grouping that warrants further investigation or intervention activity including action to mitigate or eliminate the condition represented by the physical characteristic of the reported event. As a result, the present invention modifies the location characteristics of the reported events to more accurately reflect actual conditions. The creation of an artificial group that more closely represents actual conditions is illustrated in FIG. 1 by grouping [0026] It should be noted that while the descriptions and illustrations herein relate specifically to spatially referenced physical events occurring within an area, the invention is equally applicable to physical events associated with one another in other dimensions and occurring within a selected zone, e.g. temporal relationships within a time interval. [0027]FIG. 2 illustrates a specific exemplary application of the present invention with respect to an area of analysis [0028] Selected physical events [0029] Population density information is presently available from which the assembled sampling space or universe would include individuals particularly at risk to a certain health event. For example, the universe of sample locations may include the population density of individuals of a childhood age when determining the presence of statistically significant clustering of cases of childhood leukemia. Alternatively, the universe of possible sample locations could include the at-risk population density in view of tho probability that an individual in the general population would be exposed to a condition conducive to contracting a particular disease. [0030]FIG. 2 further schematically illustrates the steps performed, such as by computer processor [0031] As generally indicated in FIG. 2, the generation of reference distribution [0032] In general, it is convenient to express the test statistic as a cross product (a Γ product) although this invention applies generally to all statistics calculated from spatially referenced data. More specifically, for spatially referenced data, the Γ product is:
[0033] Those skilled in the art will recognize that the null hypothesis for this cross-product statistic is that observations on z are independent of proximity. The alternative hypothesis being that observations on z are in some way associated with proximity. [0034] There are three general measures of proximity that provide a ready means for quantifying spatial relationships in the Γ product. As is known in the art, these proximity measures quantify spatial and/or temporal relationships between pairs of points and are of three basic types: adjacency, distance, and nearest neighbor. Those skilled in the art will also appreciate the advantages and disadvantages of each of these three types as well as that each may be used in the method described and claimed herein. For exemplary purposes, adjacency based statistics such as join-count and Moran's I statistics may be used for adjacency based analysis whereas Mantel's test for distance-based and Cuzick and Edwards' test for nearest neighbor-based analysis may also be used. The equations for these analyses are generally recognized in the art. For completeness, Mantel's distance-based cross product statistic for space-time clustering is:
[0035] As shown in FIG. 2, spatially restricted distribution [0036] A graphic illustration of reference distribution [0037] Credibility [0038] To this point, reference has been made to randomly sampling a universe of possible sample locations such as the at-risk population within an area in order to determine sample locations for the allocation of the physical characteristics. While a multitude of options are available to generate the sample space, four specific modeling alternatives are contemplated for use with the present invention. More particularly, the point model [0039] It will be apparent to one skilled in the art from the following description, that the accuracy of the information obtained from point model [0040] Point models [0041] Population models [0042] Polygon models [0043] As described above, the point, population, and polygon models are point density functions describing the locations where events could have occurred. Risk model [0044] Various other advantages and modifications will become apparent to one skilled in the art after having the benefit of studying the teachings of the specification, the drawings, and the followings claims. Referenced by
Classifications
Legal Events
Rotate |