Difference between revisions of "TM Merge - Algorithm Settings"

Revision as of 09:36, 3 August 2018

Algorithm

The Aquila/TM Merge program is delivered with some sensible defaults for matching patients, however it's fairly inevitable that some changes will need to be made.

Algorithm Process

• Incoming patient records (from Telemarque) are given a quality score based on which identifiers and key demographic information each record possesses, and in some cases whether images or Telemarque Assessments are present in the Telemarque data.
• Based on this score, a decision is made whether to import or not. (Empty records, with little demographic information and no images etc are not imported)
• Where a record is okay to import, a further decision is made whether the demographics quality is sufficient to apply the matching algorithm.
• If the demographics quality is sufficient to import, but not at the required standard for running the algorithm, the record is imported as a new patient into Aquila
• If the Telemarque record quality is good, then the program will use the algorithm to attempt to match patients between the two systems.
• Note that the algorithm will only attempt to match data items if that data item is present in both the TM and the Aquila patient records. There is no match score penalty for missing data.

Interpreting the Results

The spreadsheet will show:

• Results Summary Page: how many patients imported, merged, failed etc plus the algorithm settings (for back reference) & the time of run, and algorithm version
• Exception Log: this should be zero in all cases - any exceptions should be reported
• Full Patient Data. This is where to look for patient matches:

The "Full Patient Data" sheet contains the following:

• the incoming Telemarque patient details in the first (left most) columns
• the following columns show the result of the merge (e.g. "Import as new patient", "Matched Patient & Above Confidence Threshold", etc)
• the Aquila Patient ID and details that it has found (any in italics did not meet the full requirements for a match, but could be considered "close)
• the Algorithm details - count of matches, scores for matching, required score, and the final column is a detailed report of the algorithms progress.

The key thing we need to avoid is FALSE POSITIVE MATCHES because unpicking two different patients in AquilaCRS that the merge has combined together is likely to be a difficult and error prone task

Far better to allow the merge to create new patients as it will be easier in AquilaCRS to combine two patient records together (assuming they're the same person of course).

Adjusting the Algorithm

BEFORE adjusting the algorithm settings please consider whether it would be more prudent to update the demographics data either in Aquila or in the Patient Master File.

If you see a close match in the spreadsheet that you consider to be the same person, you should investigate which aspect of the demographics it failed to match, and then update whichever database is the one missing the data.

For example: if there's a close match between two patients that is clearly the same person, then either Aquila or the Telemarque (Patient Master File) will have some mismatched or missing data. Correct the data in the appropriate system and re-run the merge.

If, after adjusting data in either system a record is stil not merged then you can consider one of the following strategum:

a) Consider increasing the relative importance of the demographics item in question (i.e. give the Date of Birth a greater importance)

b) Consider lowering the required confidence score as a whole, so that the score given to the close match record now passes.

CAUTION : Altering the algorithm scores will affect the algorithm results for ALL records - after each alteration, very careful attention should be made to ensure that there are no false positives.

Required Confidence Score to Merge Patients: This is the score that the match has to achieve in order to merge the TM patient with the Aquila Patient

Individual Confidence Scores: The algorithm calculates a "match score" for each patient in Aquila when compared to the incoming Telemarque patient. Each matching element (NHS Number, DoB, etc) is given a score based on the values in this section. Note that after significantly changing any of these scores, you should review the Required Confidence Score as it will have changed relative to the total available score

Include Phonetics in Aquila Search: In order to process Aquila patients using the Algorithm, first a subset of Aquila patient records must be collected. This option will include Aquila patients using phonetic matching for names and will increase the size of the subset. In the Algorithm, if either the first or last name matches on the phonetic code, the score will be increased by only 40% of the individual score)

Allow Name Variants in Algorithm: Applies to first (given) name only. Allows shortened versions of first names to count as a match or not. Will only score 60% of the First Name score if a name variant matches.

Fuzzy match text, ID strings & DoB: Applies to NHS Number, Given Name, Family Name, Address line 1 and Town. If ticked, the algorithm will match the data if the edit distance between the two data items is 1. That allows for a single replaced or omitted character, or a single transposition of two characters. Matching values score only 60% of the individual score value.#

Penalise Mismatching Data Points: When this setting is on, any data items that do not match using any available method have the match score reduced by the Penalty percentage. This defaults to 25%. For example, a patient is found in the subset query and the NHS Number does not match the incoming Telemarque patient, then the match score is penalised by 25% of the NHS Number individual score which defaults to 90 (and thus we subtract 15 from the match score). When this setting is off - the match score is not affected by mismatched data items.

Use extended address in Algorithm: this setting allows the inclusion or exclusion of the address line 1 and town in the algorithm.

SAVE: If you wish to save the Algorithm settings for next time, press the Save button. Otherwise your settings will not be saved.