Database reduction based on geographically clustered data to provide record selection for clinical trials

Abstract

Aspects and features relate to computationally reducing the size or complexity of such databases to improve the speed and efficiency with which such databases are processed by a computing system to identify investigators for clinical trials. In some aspects, a processing device performs operations including identifying a data source for geographically clustered data containing corresponding descriptors for database records. The operations further include formatting the corresponding descriptors to generate standardized corresponding descriptors and matching each standardized corresponding descriptor to generate a record score for the descriptor. The record scores can be combined to generate an overall score for each database record, and the database records can be selected and written to a data store based on the overall score.

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims priority to and claims the benefit of U.S. Provisional Application No. 62 / 833,328, filed April 12, 2019, which is incorporated by reference in its entirety.

[0002] The present disclosure relates generally to data processing techniques for eliminating duplicate, inconsistent, or otherwise erroneous records in databases. More specifically, but not by way of limitation, the present disclosure relates to using such techniques to reduce the size or complexity of such databases to improve the speed and efficiency with which the databases are processed by computing systems to identify investigators for clinical trials based on geographic and clinical performance considerations. [Background technology]

[0003] Clinical trials are an important part of the process of introducing new therapies into the health care system. Such new therapies may include new vaccines, compositions (e.g., pharmaceutical compositions), dietary supplements, medical and / or dietary choices, and / or medical devices. Clinical trials may be utilized to generate data regarding safety, efficacy, patient compliance, ease of use, and other items related to the treatment. Clinical trials may vary in size and cost, and they may involve a single research center or multiple research centers in a single country or multiple countries.

[0004] Over time, physicians practicing medicine choose to be trained to participate in trials as clinical trial investigators. Investigators may change jobs, get married, etc., resulting in seemingly different investigator entries in the clinical trial investigator database. The clinical trial investigator database grows in size and complexity as investigator details are added. The combination of information from several sources over time through mergers and acquisitions perpetuates the situation, creating a database that contains multiple instances of the same individual over time. Database queries may return an overestimate of the number of investigators previously employed, and experience values ​​for a particular investigator may be underestimated across multiple entries. Manual review of database entries is time consuming and error prone, especially when the task is divided across more than one individual. Summary of the Invention [Means for solving the problem]

[0005] In one embodiment, the system includes a data storage device, a non-transitory computer readable medium including computer program code, and a processing device communicatively coupled to the data storage device and the non-transitory computer readable medium. The processing device is configured to execute the computer program code and perform operations. The operations include identifying a data source for the geographically clustered data containing corresponding descriptors for database records to be written to the data storage device. The operations further include formatting the corresponding descriptors to generate standardized corresponding descriptors, and matching each standardized corresponding descriptor to generate a record score for the descriptor. The operations further include combining the record scores for the standardized corresponding descriptors to generate an overall score for each database record, and selectively writing each database record to the data storage device based on the overall score to compile the database.

[0006] In another embodiment, the method includes identifying a data source for the geographically clustered data containing corresponding descriptors for database records to be written to a data store. The method further includes formatting the corresponding descriptors to generate standardized corresponding descriptors, and matching each standardized corresponding descriptor to generate a record score for the descriptor. The method further includes combining the record scores for the standardized corresponding descriptors to generate an overall score for each database record, and selectively writing each database record to the data store based on the overall score to compile the database.

[0007] In another embodiment, a non-transitory computer readable medium includes computer program code executable by a processor to cause the processor to perform operations. The operations include identifying a data source for the geographically clustered data containing corresponding descriptors for database records to be written to a data storage device. The operations further include formatting the corresponding descriptors to generate standardized corresponding descriptors, and matching each standardized corresponding descriptor to generate a record score for the descriptor. The operations further include combining the record scores for the standardized corresponding descriptors to generate an overall score for each database record, and selectively writing each database record to the data storage device based on the overall score to compile the database.

[0008] In some examples, each database record corresponds to a clinical trial investigator, a clinical trial site corresponding to the clinical trial investigator, or both, and the operations further include using spatial cluster analysis to generate geographically clustered data to determine the number of prospective clinical trial subjects within a defined distance of the prospective clinical trial investigator. The present invention provides, for example, the following items. (Item 1) 1. A system comprising: A data storage device; a non-transitory computer readable medium containing computer program code for database record selection; a processing device communicatively coupled to the data storage device and the non-transitory computer readable medium, the processing device executing the computer program code; identifying a data source for the geographically clustered data containing corresponding descriptors for database records to be written to the data store; formatting the corresponding descriptor to generate a standardized corresponding descriptor; matching each of the standardized corresponding descriptors and generating a score for each standardized corresponding descriptor; combining the record scores for the standardized corresponding descriptors to generate an overall score for each of the database records; and selectively writing each database record to said data storage device based on said overall score to compile a database; a processing device configured to perform operations including: A system comprising: (Item 2) The recorded score is generating a Damerau-Levenshtein score from the matching to generate a modified Damerau-Levenshtein score; and generating a binary score using the modified Damerau-Levenshtein score; 2. The system of claim 1, generated by an operation including: (Item 3) 3. The system of claim 2, wherein the overall score is generated using a machine learning model trained using known matches and mismatches. (Item 4) 2. The system of claim 1, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to the clinical trial investigator. (Item 5) 5. The system of claim 4, wherein the operations further include using spatial cluster analysis to generate the geographically clustered data and determine a number of prospective clinical trial subjects within a defined distance of the clinical trial investigator from among a plurality of prospective clinical trial investigators. (Item 6) The operation further comprises: determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and 6. The system according to item 5, comprising: (Item 7) 7. The system of claim 6, wherein the operations further include updating the data store based on clinical trial information. (Item 8) 1. A method comprising: identifying, by a processing device, a data source for the geographically clustered data containing corresponding descriptors for database records to be written to a data storage device; formatting, by the processing device, the corresponding descriptor to generate a standardized corresponding descriptor; matching, by the processing device, each of the standardized corresponding descriptors to generate a recorded score for each standardized corresponding descriptor; combining, by the processing device, the record scores for the standardized corresponding descriptors to generate an overall score for each of the database records; selectively writing, by said processing device, each database record to said data storage device based on said overall score to compile a database; A method comprising: (Item 9) The matching of each standardized corresponding descriptor is further generating a Damerau-Levenshtein score from the matching to generate a modified Damerau-Levenshtein score; and generating a binary score using the modified Damerau-Levenshtein score; The method according to item 8, comprising: (Item 10) 10. The method of claim 9, wherein the overall score is generated using a machine learning model trained using known matches and mismatches. (Item 11) 9. The method of claim 8, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to said clinical trial investigator. (Item 12) 12. The method of claim 11, further comprising using spatial cluster analysis to generate the geographically clustered data and determine the number of prospective clinical trial subjects within a defined distance of the clinical trial investigator from among a plurality of prospective clinical trial investigators. (Item 13) determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and Item 13. The method of item 12, further comprising: (Item 14) 14. The method of claim 13, further comprising updating the data store based on clinical trial information. (Item 15) A non-transitory computer readable medium comprising computer program code executable by a processor to cause the processor to perform operations, the operations comprising: identifying a data source for the geographically clustered data containing corresponding descriptors for database records to be written to a data storage device; formatting the corresponding descriptor to generate a standardized corresponding descriptor; matching each of the standardized corresponding descriptors and generating a score for each standardized corresponding descriptor; combining the record scores for the standardized corresponding descriptors to generate an overall score for each of the database records; and selectively writing each database record to said data storage device based on said overall score to compile a database; 16. A non-transitory computer readable medium comprising: (Item 16) The recorded score is generating a Damerau-Levenshtein score from the matching to generate a modified Damerau-Levenshtein score; and generating a binary score using the modified Damerau-Levenshtein score; Item 16. The non-transitory computer-readable medium of item 15, generated by an operation including: (Item 17) 17. The non-transitory computer-readable medium of claim 16, wherein the overall score is generated using a machine learning model trained using known matches and mismatches. (Item 18) 16. The non-transitory computer-readable medium of claim 15, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to the clinical trial investigator, and the operations further include using spatial cluster analysis to generate the geographically clustered data and determine a number of prospective clinical trial subjects within a defined distance of a clinical trial investigator from among a plurality of prospective clinical trial investigators. (Item 19) The operation further comprises: determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and 20. The non-transitory computer-readable medium of item 18, comprising: (Item 20) 20. The non-transitory computer-readable medium of claim 19, wherein the operations further include updating the data store based on clinical trial information. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram depicting a system for providing database record selection for clinical trials according to an aspect of the present disclosure.

[0010] [Diagram 2] FIG. 2 is a software entity flow diagram depicting a system for providing database record selection for clinical trials in accordance with an aspect of the present disclosure.

[0011] [Diagram 3] FIG. 3 is a flow chart illustrating a database reduction process used in a system for database record selection according to an aspect of the present disclosure.

[0012] [Figure 4] FIG. 4 is a flow chart illustrating a process for compiling and using a clinical investigator database for a clinical trial according to an aspect of the present disclosure.

[0013] [Diagram 5] FIG. 5 is a logic diagram illustrating an example of a data source matching logic flow that may be used in database reduction according to an aspect of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Aspects and features of the present disclosure provide a system that uses computer algorithms to determine similarity of database entries, reducing the manual effort of quality control checks, adjudication of borderline cases, or both. These algorithms can be triggered automatically as new data is added, thus maintaining database quality and keeping database size as small as possible. These optimizations ensure faster and more resource-efficient processing of data when computerized selection of clinical trial investigators is employed.

[0015] In some embodiments, the system identifies a source for a data database containing corresponding descriptors for records to be written to a data store. The data store includes a database for use in selecting investigators for clinical trials. The data store can also include other databases, including databases for use in selecting subjects for clinical trials. In some embodiments, the system also formats the corresponding descriptors for consistency, matches standardized corresponding descriptors, and generates a record score. The record scores can be combined to generate an overall score for each database record, and database records may or may not be written (selectively written) based on the score to compile a reduced-size database of clinical trial investigators.

[0016] The matching described above can, in some aspects, be performed using the Damerau-Levenshtein method to generate a Damerau-Levenshtein score. The Damerau-Levenshtein score can then be used to provide a binary score that determines whether a database record is used in the clinical trial investigator's database. Machine learning models trained using known matches and mismatches can be used to improve the efficiency and accuracy of matching. Adaptive feasibility can be used to fine-tune clinical trial investigator selection. Additionally, once a clinical trial is underway, the data store described above can be updated based on clinical trial information.

[0017] Detailed descriptions of certain embodiments are discussed below. These illustrative examples are provided to introduce the reader to the general subject matter discussed herein and are not intended to limit the scope of the disclosed concepts. The following sections describe various additional aspects and embodiments with reference to the drawings, in which like numbers indicate like elements, but the directional descriptions are used to describe the illustrative embodiments and should not be used to limit the disclosure as such.

[0018] Referring now to the drawings, FIG. 1 depicts an example of a system 100 for providing clinical trial investigator and subject selection, according to some aspects. FIG. 1 depicts an example of hardware components of the system 100. The system 100 includes a statistical computation environment server (computation server) 102. The computation server 102 may be a computer or other machine that provides the cluster analysis, modeling, and calculations required for database reduction. The computation server 102 may include one or more other systems. For example, the computation server 102 may include an adapter, router, etc. for accessing a communication network. In this example, the computation server 102 is connected to a data network 104. The data network 104 may also be fully embedded within (or may include) the Internet, an intranet, an extranet, or a combination thereof. In one example, communication between two or more systems or devices may be accomplished by a secure communication protocol, such as Secure Sockets Layer ("SSL") or Transport Layer Security ("TLS"). The system 100 includes a database server 106, which is connected to the computation server 102 through the network 104. Database server 106 is connected to data storage device 107, which may include a variety of databases, for example, databases corresponding to or including information about clinical trial investigators, clinical trial subjects, or other clinical trial information.

[0019] Still referring to FIG. 1 , the calculation server 102 further includes a processing device 108 communicatively coupled to a non-transient memory device 110. The non-transient memory device 110 includes a computer-readable medium used to store computer program code 112 for causing the processing device 108 to perform operations for database reduction, database record selection, and assistance functions according to aspects of the present disclosure. The non-transient memory device 110 may also include cached files 114, such as database records, currently being used by the processing device to perform these operations. The processing device 108 may include a processor or multiple processors, with non-limiting examples of processing devices including field programmable gate arrays ("FPGAs"), application specific integrated circuits ("ASICs"), microprocessors, and the like.

[0020] Continuing with FIG. 1, the data network 104 in the system 100 connects the computation server 102 to the client computing devices 118 and 124. The computing device 118 is a mobile device by which a clinical trial subject can access information about the clinical trial, for example, via an app or the web. The computing device 124 is a computer system that a clinical trial investigator or personnel associated with a clinical trial investigator can use to access information about the clinical trial. The number of devices depicted in FIG. 1 is provided for illustrative purposes. A different number of devices may be used. For example, although each device, server, and system in FIG. 1 is shown as a single device, multiple devices may instead be used. The data network 104 may include one or more of a variety of different types of networks, including a wireless network, a wired network, or a combination of wired and wireless networks.

[0021] The memory device 110 can store, at least in part, computer program code or instructions executable by the processing device and retain such stored information when power is turned off. The non-transitory computer readable media contained within the memory device 110 can include electronic, optical, magnetic, or other storage devices capable of providing computer readable instructions or other program code to a processing device. Such media may store instructions on an installation server prior to installation in or on the server depicted in FIG. 1. Non-limiting examples of computer readable media include (but are not limited to) magnetic disks, memory chips, read only memory (ROM), random access memory ("RAM"), ASICs, configured processing devices, optical storage devices, or any other media from which a computer processing device may read instructions.

[0022] FIG. 2 is a software entity flow diagram depicting a software system 200 for providing database record selection and other functions for a clinical trial according to an aspect of the disclosure. Generally, the software entities on the left side of FIG. 2 are used pre-trial, as indicated by the lower adjacent arrow. After the clinical trial starts 201, the software entities on the right side of FIG. 2 manage the clinical trial. Software block 202 provides study protocol design services, including modeling to determine inclusion and exclusion criteria. To study the appropriate disease state patient population and level of diversity, the investigator defines criteria that determine whether a patient is eligible for the trial. Inclusion and exclusion criteria can encompass disease and treatment specific characteristics, including patient characteristics (e.g., age, genetic profile) and previous laboratory test results related to the disease and / or condition. An additional parameter is the number of subjects required for the clinical trial. Clinical trial parameters may further include the desired timing for enrolling subjects and / or investigator sites and a proposed timeline for completing the clinical trial.

[0023] Block 204 of the system 200 provides cluster analysis to identify geographic clusters of subjects around a clinical investigator based on a defined distance and to identify additional investigators for clusters of subjects that are too far from a known or requested investigator. The cluster analysis utilizes a database 206 of prospective clinical trial subjects, which includes information regarding the admission of subjects admitted to clinical studies, past results, and clinical information. The cluster analysis also utilizes a database 208 of clinical trial investigators and clinical trial sites. These databases can reside in the data store 107. A longitudinal analysis 210 is applied to the data from the database 206 to provide a historical perspective on the data. The longitudinal analysis includes data regarding the last sample used from a subject and data regarding analyte levels over time for prospective study subjects.

[0024] The site scoring block 211 provides a ranking of the study sites based on the quality of the investigator as determined based on historical performance. Throughout this disclosure, the term "investigator" may be used synonymously with the term "site" when discussing clinical trials. A site may be a particular laboratory or research institute that corresponds to an investigator or with which a defined investigator is affiliated.

[0025] For spatial clustering, each possible object is cn(k) where c=1 and n=1 identify a first investigator location in a first country with k possible subjects within a defined distance (k depends on c and n), c=1 and n=2 identify a second possible investigator location in the first country, etc., for C countries and n=1,..N(c) investigator locations with k(nc) possible subjects, and so on. The spatial or "geographical" clusters as determined by the cluster analysis block 204 are defined as follows: c This can be expressed by minimizing the number of [ka] where p is set at a level corresponding to an acceptable level of certainty of meeting the study timeline and the actual conversion of the subject to a clinical trial patient. The location information may include latitude and longitude. GPS data, zip code, physical address, and / or postal code may be used to determine the latitude and longitude. The clinical trial parameters are used to query a database to determine potential subjects and investigators for the clinical trial. The database provides information related to each potential subject, including, but not limited to, the geographic location of the subject. Similarly, the database provides investigator information related to each potential investigator, including, but not limited to, the geographic location of the investigator. In some examples, if there are clinical investigators with insufficient experience, the cluster analysis 204 can provide a list of potential investigators that could be used if trained. For example, if the study subject is also a physician but has no experience as a clinical investigator, training a physician to be an investigator for the current clinical trial may be considered.

[0026] Continuing with FIG. 2, software block 212 identifies study sites (investigators) from the clustered data. To make this process as computationally efficient as possible, software block 212 also includes a database reduction technique in accordance with aspects described herein. This database reduction technique is described in detail below with respect to the remaining figures. Site identification block 212 uses a country-site distribution engine to determine a global distribution and a mixed country-site set for the clinical trial. Site identification can include collecting, for each country in the country-site distribution, the number of bids for the clinical trial based on historical recruitment rates and based on regulatory time requirements. Software block 212 can also provide graphical tools that list previously employed investigators and sites, their ranking based on historical performance, and their geographic distribution. Data regarding observed country-level screening rates can also be displayed.

[0027] During the pre-trial phase for a typical clinical study, potential investigators and subjects are contacted and responses are collected. Initial contact results in responses that may change the selected distribution of subjects, study sites, clinical investigators, or other parameters. The adaptive feasibility software block 214 updates all of this information as the final responses modify the optimal geographic clustering and distribution of subjects and investigators.

[0028] The global trial site distribution software block 216 provides pre-award and post-award modeling. This modeling software tool provides trial site distribution based on country-specific recruitment rates and requirements. It provides time frames for clinical trials and requirements based on those time frames. Block 216 also includes a country-site mix calculator. The country-site mix and other determinations can be updated during the trial. The software block 216 manages the clinical study both before and after the study is initiated. The trial site distribution software performs database record selection to populate the data to perform the calculations. These calculations are made faster and more efficient than would otherwise be possible by ensuring that the data in the database is correct, i.e., consistently formatted, without duplications caused by issues such as changes in the spelling of names or changes in biographical information over time. This process may be referred to herein as data reduction, data deduplication, or database reduction.

[0029] Still referring to Figure 2, software block 218 provides study information and data extraction and feedback for both site identification, subject identification, and both into database 206 so that the database is kept up to date for future clinical trials. Study management software block 220 provides benchmarking, communication, and appropriate dashboards for interaction with both subjects and investigators. Information from these dashboards can be displayed and input can be received through client computing devices 118 and 124, as an example.

[0030] Software entities within the system 200 incorporate real-world clinical data (including laboratory diagnostic test results) and allow users to assess the impact of protocol inclusion / exclusion criteria on the pool of available people with the disease, with the potential to design more efficient trials. Clustering uses the proximity of patients and best-suited trial investigators to identify investigators with a greater density of available patients nearby. It also assists in planning the number of countries and investigators required for a trial. The user interface provides the ability to run complex mathematical and statistical procedures without technical knowledge, within a controlled environment that prevents misapplication.

[0031] The system utilizes robust statistical and mathematical algorithms that are run within the statistical computing environment server 102. Linear programming techniques are applied to determine potential country-site distributions that achieve study timelines based on the most recent regulatory submission timelines within each country and screening rates from previously conducted similar studies. In some embodiments, the statistical computing environment can be established using a statistical programming language such as R to provide filtering, ordering, grouping, summarization, and visualization of the data.

[0032] FIG. 3 is a flow chart illustrating a process 300 for database reduction according to an aspect of the disclosure. The process 300 provides for compiling a list of unique clinical trial investigators from sources of information that are not standardized (no spelling checks and no attempt to make the format of the entries consistent either between or within the data sources). Common components of the pieces of information contained in each source are identified and compared between sources, generating a similarity score for each component. The similarity scores for each component are then combined to give an overall similarity score for the clinical trial investigator entries for each pairwise comparison of investigators both within and between sources. The identification of values ​​(classifiers) for the scores that define a binary outcome ("Match", "No Match") allows the process to function efficiently.

[0033] At block 302 of process 300, internal and external sources of clinical investigator information are identified. At block 304, the clinical investigator information is checked to make the entries consistent in format. At block 306, the database records from the different sources are compared and unique clinical investigators across the different sources of clinical investigator information are identified. At block 308, the database records are written to a data store to compile a clinical investigator database that is used in determining global study site distribution and organizing clinical trials. Reduction in the size of the database may be referred to herein as de-duplication or reduction, since duplicate database entries are automatically identified and eliminated, which reduces the size of the database.

[0034] FIG. 4 is a flow chart illustrating an example of a process 400 for compiling and using a database of clinical investigators for clinical trials according to an aspect of the disclosure. In block 402, geographically clustered data is generated by the processing device 108 using cluster analysis 204. The geographically clustered data uses cluster analysis to determine the number of prospective clinical trial subjects within a defined distance of the prospective clinical trial investigator. In block 404, sources for the geographically clustered data that contain corresponding descriptors for the database records are identified. Data sources that contain similar descriptors for the entries being databased are identified; while the number of selected descriptors needs to be the minimum necessary, the set of descriptors needs to be a sufficient subset of all potential descriptors such that they collectively and uniquely identify the entries if complete information exists.

[0035] In some examples, the external source can be assumed to be the most accurate source in terms of unique entries. This source is external to the company designing the clinical trial such that a database reduction process would otherwise be unnecessary. For the purposes of this example, it may be assumed that there are three internal sources of data that need to be combined and deduplicated, and one external source of data, which is assumed to be a deduplicated database.

[0036] In block 406 of FIG. 4, the processing device 108 formats the corresponding descriptors and generates standardized corresponding descriptors. For the purposes of this example, it may be assumed that there are five descriptors of interest per database entry in each of the selected sources: Investigator Name, Investigator Facility Street Address, Investigator City, Investigator Zip Code, and Investigator Email Address. In this example, the Investigator Name includes a combined first and last name. For Southeast Asian names, matching may be applied twice, once to reverse the order of names in a database entry to accommodate the cultural norm that a person's family name is mentioned first. The Investigator Facility Street Address, in this example, includes the building / block number as a mandatory part for definitive identification of the location, since there are scenarios where the street alone would not be sufficient. For example, Harley Street in London, England is world-renowned as a street of medical practitioners, and the use of only Harley Street in a database entry would not distinguish between two investigators with the same name who both work there. Each descriptor needs to be standardized in format as much as possible to facilitate the matching process. Standardization that may be incorporated into the process includes, as examples, converting all alphabetic characters to lower case and removing punctuation marks.

[0037] Still referring to FIG. 4, in block 408, each standardized corresponding descriptor is matched to generate a record score for the standardized corresponding descriptor. In some embodiments, a machine learning model is used. In some embodiments, the Damerau-Levenshtein (DL) method is used for text matching. Each descriptor undergoes a text matching process separately, both within a data source entry and between data sources. As an example, consider four names: Stephen, Steve, Steven, and Stephanie. The DL algorithm determines the number of insertions, deletions, and character transpositions required to obtain one name from the other. For example, the difference between Steve and Steven generates a DL score of 1, since the change only requires the addition of the letter "n". The DL score based on the difference between Stephanie and Stephen is 3 (the addition of "i" and "e" and the change from "a" to "e"). The difference between Stephanie and Steve produces a DL score of 5 (removal of "h", "a", "n", and "i" and substitution of "p" and "v").

[0038] The length of the text being matched may be important. For example, the difference between "dog" and "cat" would generate a DL score of 3, which is the highest value for a three-word word. For this reason, the DL score is modified by dividing the DL score by the number of characters in the longer of the two words being compared to generate a modified DL score. Using the example based on "Steve" and "Steven" above, the difference would be scored as 1 / 6, and the difference between "Stephanie" and "Steve" would be scored as 5 / 9. The final modification is to subtract the modified score from 1 and round to generate a binary final recorded score from the modified DL score, with a score of 1 representing a match and a score of 0 representing a perfect mismatch.

[0039] For street address descriptors, a correction is needed if the street number is considered important. As an example, the system can be designed such that the numerical portions of the descriptors being compared must match exactly, otherwise a record score of 0 is assigned to the record. With any of the matching described above, to avoid false mismatches and therefore falsely creating duplicates, a query to the user is invoked if there is a good match on all descriptors except for the concrete portion such as the street number.

[0040] Continuing with FIG. 4, in block 410, the record scores are combined to generate an overall score for the database record including the corresponding standardized descriptors. In one embodiment, once all the descriptors have been scored, a k-means machine learning approach can be used to combine the record scores for the corresponding standardized descriptors and generate an overall score for the database record. A training set of known matches and mismatches can be used to train a machine learning model that is incorporated into the process using k=2 (match='yes' or 'no'). In block 412, the database record is written to the database 208 in the data storage device 107. The database record may or may not be written (selectively written) based on the overall match score corresponding to 'yes' or 'no' as explained immediately above, with 'yes' dictating that the record becomes part of the clinical investigator's compiled database. Records are written to compile the database until all records have been processed. In block 414, the processing device 108 determines whether there are any remaining records to be processed. If so, the process is repeated for each record from block 406. Otherwise, the compiled database will be used. Adaptive feasibility is used to generate a test site distribution in block 416.

[0041] FIG. 5 is a logic diagram illustrating an example of a data source matching logic flow 500 that may be used in database reduction according to aspects of the disclosure. Database reduction begins by matching each internal data source 502 with at least one external data source 504. If no match exists at decision block 506, then an external data source 508 is matched. In this example, there are three external data sources: external data source A, external data source B, and external data source C. If a match can be made between the internal data source and the external data source at block 506, then a unique ID is assigned to the matched record at block 510. Similarly, if a match can be made between the internal data source 508 at decision block 512, then a unique ID is assigned to the matched record, again at block 510. For records that remain unmatched from the internal data source 508, the system may create a unique ID at block 516. The ID created for the unmatched record can be used to identify the record for future reference.

[0042] Unless specifically stated otherwise, throughout this specification, terms such as "processing," "computing," "calculating," "determining," or the like refer to the actions or processes of a computing or processing device, such as one or more computers or similar electronic computing devices, that manipulate or transform data, which is represented as physical electronic or magnetic quantities in the memory, registers, or other information storage, transmission, or display devices of a computing platform.

[0043] The system or systems discussed herein are not limited to any particular hardware architecture or configuration. A computing device may include any suitable arrangement of components that provide results conditioned on one or more inputs. Suitable computing devices include general-purpose microprocessor-based computing systems that access stored software that programs or configures the computing system, from general-purpose computing devices to specialized computing devices that implement one or more aspects of the present subject matter. Any suitable programming, scripting, or other type of language, or combination of languages, may be used to implement the teachings contained herein in software to be used in programming or configuring a computing device.

[0044] Aspects of the methods disclosed herein may be implemented in the operation of such a computing device. The order of some of the blocks presented in the above examples may be varied, e.g., blocks may be rearranged, combined, or divided into sub-blocks. Certain blocks or processes may be performed in parallel.

[0045] The use of "configured to" is meant herein as an open-ended and inclusive term that does not exclude devices configured to perform additional tasks or steps. Additionally, the use of "based on" refers to an action or process of a computing or processing device, and is meant to be open-ended and inclusive in that a process, step, calculation, or other action "based on" one or more recited conditions or values ​​may, in practice, also be based on additional conditions or values ​​other than those recited. The headings, lists, and numbering contained herein are for ease of discussion only and are not meant to be limiting.

[0046] The foregoing description of embodiments, including illustrated embodiments of the subject matter, has been presented for illustrative and explanatory purposes only and is not intended to be exhaustive or to limit the subject matter to the precise form disclosed. Numerous modifications, adaptations, and uses thereof will be apparent to those skilled in the art without departing from the scope of the present subject matter. The illustrative examples described above are provided to introduce the reader to the general subject matter discussed herein and are not intended to limit the scope of the disclosed concepts.

Claims

1. 1. A system comprising: A data storage device; a non-transitory computer readable recording medium having stored thereon computer program code for database record selection; a processing device communicatively coupled to the data storage device and the non-transitory computer-readable recording medium, the processing device executing the computer program code; identifying a plurality of database records to be written to the data storage device, the plurality of database records being contained within a data source for geographically clustered data, each database record of the plurality of database records including a plurality of texts, each of the plurality of texts corresponding to a respective one of a plurality of descriptors; formatting the plurality of texts to generate a plurality of formatted texts, each formatted text corresponding to a respective standardized descriptor of a plurality of standardized descriptors; performing a pairwise comparison of every two database records of the plurality of database records, where performing the pairwise comparison includes matching the formatted text corresponding to each standardized descriptor of each database record of the plurality of database records with the formatted text corresponding to the same standardized descriptor of every other database record of the plurality of database records, and generating a record score for each standardized descriptor for each pairwise comparison of two database records of the plurality of database records, where the record score comprises: generating a Damerau-Levenshtein score from the match and generating a modified Damerau-Levenshtein score by dividing the Damerau-Levenshtein score by the number of characters in the longer of the two formatted texts being matched; generating a binary score by subtracting the modified Damerau-Levenshtein score from 1 and rounding, the binary score being the record score, the record score indicating the degree of matching; and for each pairwise comparison of two database records of the plurality of database records, outputting an overall score by inputting the record scores generated for each of the standardized descriptors into a machine learning model, the machine learning model being trained using known matches and mismatches; selectively writing each database record to said data storage device based on said overall score to compile a database; a processing device configured to perform operations including: A system comprising:

2. 10. The system of claim 1, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to said clinical trial investigator.

3. 3. The system of claim 2, wherein the operations further include using spatial cluster analysis to generate the geographically clustered data and determine a number of prospective clinical trial subjects within a defined distance of the clinical trial investigator from among a plurality of prospective clinical trial investigators.

4. The operation further comprises: determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and The system of claim 3 , comprising:

5. The system of claim 4 , wherein the operations further comprise updating the database stored in the data storage device based on clinical trial information.

6. 1. A method comprising: a processing device identifying a plurality of database records to be written to a data storage device, the plurality of database records being contained within a data source for geographically clustered data, each database record of the plurality of database records including a plurality of texts, each of the plurality of texts corresponding to a respective one of a plurality of descriptors; the processing device formatting the plurality of texts to generate a plurality of formatted texts, each formatted text corresponding to a respective standardized descriptor of a plurality of standardized descriptors; performing a pairwise comparison for every two database records of the plurality of database records, wherein performing the pairwise comparison includes the processing device matching the formatted text corresponding to each standardized descriptor of each database record of the plurality of database records with the formatted text corresponding to the same standardized descriptor of every other database record of the plurality of database records, and generating a record score for each standardized descriptor for each pairwise comparison of two database records of the plurality of database records, wherein matching the formatted text further includes: generating a Damerau-Levenshtein score from the match and generating a modified Damerau-Levenshtein score by dividing the Damerau-Levenshtein score by the number of characters in the longer of the two formatted texts being matched; generating a binary score by subtracting the modified Damerau-Levenshtein score from 1 and rounding, the binary score being the record score, the record score indicating the degree of matching; and for each pairwise comparison of two database records of the plurality of database records, the processing device outputs an overall score by inputting the record scores generated for each of the standardized descriptors into a machine learning model, the machine learning model being trained using known matches and mismatches; said processing device selectively writing each database record to said data storage device based on said overall score to compile a database; A method comprising:

7. 7. The method of claim 6, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to said clinical trial investigator.

8. 8. The method of claim 7, further comprising using spatial cluster analysis to generate the geographically clustered data and determine a number of prospective clinical trial subjects within a defined distance of the clinical trial investigator from among a plurality of prospective clinical trial investigators.

9. determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and The method of claim 8 further comprising:

10. 10. The method of claim 9, further comprising updating the database stored in the data storage device based on clinical trial information.

11. A non-transitory computer-readable recording medium having stored thereon computer program code executable by a processor to cause the processor to perform operations, the operations comprising: identifying a plurality of database records to be written to a data storage device, the plurality of database records being contained within a data source for geographically clustered data, each database record of the plurality of database records including a plurality of texts, each of the plurality of texts corresponding to a respective one of a plurality of descriptors; formatting the plurality of texts to generate a plurality of formatted texts, each formatted text corresponding to a respective standardized descriptor of a plurality of standardized descriptors; performing a pairwise comparison of every two database records of the plurality of database records, where performing the pairwise comparison includes matching the formatted text corresponding to each standardized descriptor of each database record of the plurality of database records with the formatted text corresponding to the same standardized descriptor of every other database record of the plurality of database records, and generating a record score for each standardized descriptor for each pairwise comparison of two database records of the plurality of database records, where the record score comprises: generating a Damerau-Levenshtein score from the match and generating a modified Damerau-Levenshtein score by dividing the Damerau-Levenshtein score by the number of characters in the longer of the two formatted texts being matched; generating a binary score by subtracting the modified Damerau-Levenshtein score from 1 and rounding, the binary score being the record score, the record score indicating the degree of matching; and for each pairwise comparison of two database records of the plurality of database records, outputting an overall score by inputting the record scores generated for each of the standardized descriptors into a machine learning model, the machine learning model being trained using known matches and mismatches; selectively writing each database record to said data storage device based on said overall score to compile a database; A non-transitory computer-readable recording medium comprising:

12. 12. The non-transitory computer readable storage medium of claim 11, wherein each database record corresponds to at least one of a clinical trial investigator or a clinical trial site corresponding to the clinical trial investigator, and the operations further include using spatial cluster analysis to generate the geographically clustered data and determine a number of prospective clinical trial subjects within a defined distance of a clinical trial investigator from among a plurality of prospective clinical trial investigators.

13. The operation further comprises: determining the adaptive feasibility for said clinical trial investigator; generating a trial site distribution for the plurality of prospective clinical trial investigators based at least in part on the adaptive feasibility; and 13. The non-transitory computer readable storage medium of claim 12, comprising:

14. 14. The method of claim 13, wherein the operations further comprise updating the database stored in the data storage device based on clinical trial information.

Images